Ethan Lazuk

SEO & marketing professional.

Generative AI for SEO: “Real Life” Examples of How I’ve Used ChatGPT, Claude & Gemini for Client Work

By Ethan Lazuk

Last updated:

AI generated image representing Welcome to the Machine.

AI is everywhere, and its proliferation is only growing. 🤖

As a result, there is a constant flow of tips exploring the uses of generative AI for SEO, specifically LLMs (like in ChatGPT, Claude, and Gemini).

There are so many “uses of AI for SEO” articles, in fact, that Google didn’t even list them for me in its normal search results.

It first summarized the main themes in “From sources across the web”:

Google from sources across the web result for uses of AI for SEO query.

Each of those 21 dropdowns has 2-4 results in it. That’s a lot of blog posts.

Then we also have custom GPTs, and lots more since the GPT Store opened in January:

Custom GPT search results for SEO query.

The question is are those “ChatGPT for SEO” articles or GPTs useful in practice? 🤔

I’m not sure.

I suppose we could test them all …

🥵 But that’s a workout.

Instead, since I’ve used LLMs in my everyday SEO work, basically since early 2023, I went through my personal chat history on ChatGPT, Claude, and Gemini (Bard) to find examples of SEO work I’ve done for clients or my own site.

In other words, these are “real life” uses of AI chatbots (LLMs) for SEO based on client deliverables. 👇 🔥

I’m constantly updating this guide with new examples, as well. Some of the early ones may have older examples or not be written super well. I’ll work on that. 😁

For example, I recently got access to Gemini 1.5 Pro in the Google AI Studio with a 2 million token context window, but back when I started this guide in 2023, Gemini was still called Bard, and ChatGPT was using GPT-3.5.

I’ve also added more context about using LLMs for SEO efficiencies but not shortcuts. 🤗

This approach has long been my philosophy, but it’s even more important to mention now after Google’s latest March 2024 core update, spam update, manual actions for “pure spam,” and new spam policies, like for “scaled content abuse.”

I even noticed this guide ranking in Google for queries like:

A client wants to use generative AI to create content query in Google Search Console.

Let’s discuss. 🙌

Table of contents 🏓

Here’s what we’ll cover, including tips from each chatbot (LLM) plus related topics, like prompting tips. Again, some of these may have older examples or be written kind of blah. The information is current, though, and, above all, helpful.

I’ve put ✅’s beside the main sections for each LLM. If you’re newer to using generative AI, I’d encourage you to read the first few sections, as well.

🤖 ✍️ A big topic is using generative AI to create content. Check out the responsible use section 🕊️ for more on that, where I explained the difference between “AI-generated” and “AI-assisted” content.

Historical context on using AI chatbots for SEO purposes

Soon after ChatGPT was released to the public on November 30th, 2022, many SEOs began experimenting with how to use it for optimizations and strategy tasks.

As more generative AI chatbots became available, like Google’s Bard in March of 2023 or Anthropic’s Claude 2 in July of 2023, interest grew in their use cases. As a result, many people began creating guides for how to use LLMs for SEO.

Then came custom GPTs on November 6th, 2023. This allowed people to quickly spin up tools for specific use cases. I’ve included a list of SEO-related GPTs at the end of this article.

As articles and videos came about how to use ChatGPT (and other LLMs) for SEO, I viewed and saved the content, along with pinning any custom GPTs, and in many cases tried what was suggested.

The renaming of Bard to Gemini in February 2024 and the release of Gemini into Google Workspace as an AI assistant are among the most recent development in how AI can be used for SEO efficiencies.

We also have the advent of two new AI models, including Gemini 1.5 and Claude 3, as well as tools like Google’s NotebookLM. Other topics to watch for will be AI agents and on-device LLMs, like updates to Siri.

Responsible use of LLMs for SEO (shortcuts vs. efficiencies)

Because ChatGPT, Gemini, and other AI models are such powerful tools, the context for when and where to use LLMs for SEO is equally as important as how to use them.

My SEO philosophy centers around creating original content as part of a holistic and people-first approach.

It felt discouraging to see social media posts and videos from people encouraging other marketers or businesses to use ChatGPT and other LLMS for taking shortcuts in SEO.

I found that irresponsible.

I will say, I think the tide has turned on some of that, given recent ranking system changes by Google (like the March 2024 core update) as well as the rollout of AI Overviews, which has helped to refocus the conversation around holistic brand-focused marketing.

I’ve seen plenty of SEO content that covers AI topics.

Some of them are treasures of knowledge, while others feel like they’re created for marketing purposes themselves.

On the legit side of the coin, you have people who use what they create.

For example, Caitlin Hathaway makes custom GPTs and posts threads on X (Twitter) about how she uses them:

She also recently published an article about LLM uses for content strategy with firsthand examples.

Then there’s the other side:

Examples of people using AI to take shortcuts on TikTok.

The question is, can we say that one side is more right than the other?

People’s backgrounds, political views, and other worldviews can lead them to see situations differently.

As a user-centric SEO, I view a person who promises to “skyrocket” your website’s traffic with an “SEO boost” from AI as a bad actor who’s likely contributing spam to the internet or misleading unsuspecting site owners toward algorithmic demotions (or manual ones).

Judging from these video thumbnail images, these aren’t the types of content I’d follow. (That first web result, way down at the bottom, looks ok though.) 😉

ChatGPT Claude SEO uses SERP.

On the other hand, the people in these TikTok and YouTube video creators may see themselves as benevolent actors, like they’re helping people get ahead by accomplishing the hard work of SEO with greater ease.

I’ve met people who have that outlook. One person said they could create “thousands of pages” per day for my clients.

Here are a couple of tips for when you’re being pitched AI tools for SEO:

  1. When a salesperson shows you how their AI-generated content is ranking on Google for thousands of new keywords, or even in position 1, ask to see the traffic (many are 0 volume queries) and the direct or assisted conversions (many of the queries are irrelevant to a buyer’s journey).
  2. If the salesperson shows you a large percentage increase (a hockey stick graph) of organic traffic “skyrocketing” from their AI content or optimizations, ask to see a long-term traffic trend of 6-12+ months (often those gains are short-lived once users figure out the content is bad and Google algorithmically responds).

When it comes to using generative AI for SEO, there’s a big difference between taking “shortcuts” 🙅 and creating “efficiencies.” 🔥

SEO isn’t easy.

To become a specialist or generalist takes years of work, learning, and experience.

Generative AI can help lighten the load during that process; it can’t replace hard work or strategic wisdom. 🤗

The truth is that LLMs accentuate the knowledge and characteristics of the person using them, for better or worse.

Experienced SEOs can use LLMs to enhance their workflows and create efficiencies or even new opportunities. 🦾

Taking shortcuts with AI may achieve short-term results 📈, but these are often tenuous and can get wiped out with a Google ranking system update or after user dissatisfaction catches up algorithmically through weak engagement signals. 📉

That’s not good for brand building or your business goals.

Using LLMs to create efficiencies in SEO just means faster ways of doing the same or even higher quality work.

A quick note on “AI-generated” content

If your instinct for using ChatGPT, Gemini, or another LLM or AI tool is to generate content for SEO purposes, I’d recommend checking out the section of my helpful content guide on AI-generated content. 👈

It also covers Google’s guidance on the topic, as well as evidence of how violating those guidelines can catch up to a website’s organic traffic. 📉

In short, it’s not AI-generated content that’s an issue directly. It’s the tendency for it to be used in taking shortcuts that reduce quality and create user dissatisfaction.

That evidence has only gotten stronger since Google’s March 2024 core and spam updates.

In my experience, shortcuts taken with any kind of content creation (AI or human writers) don’t lead to the type of helpful and people-first content that satisfies readers or stands apart to achieve sustainable SEO performance.

Spammy content either has short-term or no benefits and can even hold back a site’s overall quality (the basis for Google’s old unhelpful content ML classifier). As of March 2024, the helpful content system is now incorporated into Google’s core ranking systems, which focus on page level and site-wide factors.

There are legitimate use cases for AI-generated content, though. 👀

The most common I hear is summarizing existing information into a helpful format, like creating ecommerce product page descriptions at scale. I haven’t done this myself, but I get how, when combined alongside customer reviews (UGC), it could have value.

That said, the author’s perspective (their expertise or experience) matters in many cases that users (and by extension search engines) care about, lending substance and differentiation (information gain) from other content in search results, including from AI Overviews.

Before hitting publish for your next piece of AI-generated content, ask yourself: “As someone looking for a solution to my problem, would I find this content helpful, reliable, and satisfactory?”

Another option is AI-assisted content, which means using LLMs to enhance human-generated content for a user’s benefit. I do this myself in Hamsterdam Research, using AI chatbots to help me explain esoteric AI research papers. (I’ve even included AI-generated content verbatim in some posts, like this one summarizing podcast transcripts or this one summarizing a Google’ engineer’s contributions.)

You’ll find many content-related tips throughout the rest of this guide, but they’ll always be in the context of “AI-assisted” content.

The methodology for putting this guide together

The inspiration for this guide came from reading several “X Ways to Use ChatGPT for SEO” articles and then being left wondering if the authors actually used those tips.

This isn’t a theoretical article or a comprehensive review of ways to use LLMs for SEO — it’s a list of the ways I’ve personally used ChatGPT, Claude AI, and Gemini (Bard) to accomplish SEO tasks for clients in everyday life.

However, I did catch this X thread by Ryan Law at Ahrefs that has a similar concept. I want to give him credit for having the idea first.

I’ll go through my chat history with ChatGPT Pro first, including the GPT-3.5 and GPT-4 models, including the latest version and earlier iterations with features or plugins like Browse with Bing, Advanced Data Analysis, DALL-E, and Link Reader.

Then I’ll list custom GPTs I’ve accumulated for SEO purposes.

Lastly, I’ll review my uses of Claude AI (Claude 2) and Google Gemini (Bard), including the free and Advanced versions as well as the virtual assistants in Google Workspace.

This guide was originally published in October 2023 but has continuously been updated and expanded.

How fresh are these tips? Well, the content was last updated and/or reviewed on:

Also, given the rate of advancement for generative AI technologies, some of the screenshots or tips below may apply to past iterations, designs, or brands of LLMs.

However, all the prompts and tips will work today, and I’ll do my best to change or remove any tips that no longer apply.

Also, while these tips relate to client work, some of the examples will be recreations using my website’s content or Google Search Console or GA4 data. This is to preserve the privacy of my clients’ data.

Continuing from the pretext section above, I’ll also share opinions throughout this guide about how I think generative AI should (and shouldn’t) be used for SEO.

To summarize my opinions for opening context:

  • As mentioned earlier, I believe the best approach to doing SEO today is optimizing for your users first.
  • I would never use generative AI to spin up low-quality articles just to make a quick buck or some noise on social, which also will not likely achieve lasting results for Search visibility anyway.
  • I don’t believe in using ChatGPT for doing keyword research. That’s a use case I’ve seen promoted and it doesn’t make sense because LLMs can’t generate search volume, etc. (unless you’re inputting keyword data yourself).
  • I do believe in using ChatGPT for content ideation, which I’ll share more about.
  • I don’t believe in creating AI-generated content in most instances, particularly at scale. (“AI-assisted” content is a different category.)
  • I do believe in using ChatGPT to improve your content’s quality and relevance, as we’ll explore below.

Beyond this article, I’m also exploring topics like the uses of Google Gemini in the Google AI Studio, how Perplexity AI can contribute to SEO content research, as well as reporting on AI news and developments each week in Hamsterdam.

If you have any questions while reading this article, feel free to reach out to me.

Now, let’s go through the use cases of ChatGPT for SEO. They’re followed by sections on Claude and Gemini (Bard) below.

ChatGPT uses for SEO

Based on my history with ChatGPT (GPT-3.5 to GPT-4o), here are ways I’ve used it for SEO. Some of the early ones may be older examples or a little wordy. Everything is still relevant, though:

1. Creating regex (regular expression) for Google Search Console to filter performance reports

In Google Search Console’s performance report, you can filter data by query, page, country, device, or search appearance (in addition to search type and date).

For the query and page filters, you have the option to use a regular expression filter, called Custom (regex), to show results that either do or don’t match the inputted regular expression.

Google’s own Search Console Help article (linked above) says that “Regular expression matching is tricky” and “Invalid regular expression syntax will return no matches.” The article also links to an article on regular expression syntax from GitHub.

Now, if you can comprehend the syntax of regular expressions like a native speaker, I commend you. For me, it might as well be written in hieroglyphics.

That’s why I lean on ChatGPT.

1a. Query Filtering

As a chat prompt to filter query data in GSC, I’ll write something like:

“Please create a regular expression for Google Search Console’s performance report to only return results that match the following queries.”

Then I’ll paste in a list of queries.

For this, I’ll typically just use the ChatGPT GPT-4 Default. But you can also use the free version.

Here is what the prompt looks like with a few example keywords:

ChatGPT prompt to create a regular expression for Google Search Console

The response then includes the regular expression in a box with an easy button to copy the code, as well as a breakdown explanation of how the regular expression works. This can be helpful for learning more about regex syntax as well as verifying the output matches your prompt.

Here is how the response looks:

ChatGPT response to create a regular expression for Google Search Console

The regex code looks like this:

“^(query 1| query 2| etc.)$”

And here is where to plug in the regular expression in Google Search Console:

Google Search Console Query Filter for Custom regex

And here we can see that now the query tab in the performance report only shows the 12 queries we requested data for:

Google Search Console Queries report filtered

The same can be done for pages.

There can also be variations on this.

1b. Pages filtering

To get a regex filter for GSC for specific pages, for example, I might enter a prompt to ChatGPT like:

“Please create a regular expression for Google Search Console’s performance report to only return pages that contain “hamsterdam” or “twitter” in the URL and contain any of the numbers between 11 and 25.”

Here is the prompt and response from ChatGPT, again including a breakdown, which is this time more detailed given the specificity of the regex:

ChatGPT prompt and response to create a regex for pages in Google Search Console

The regex code looks like this:

“^(?=.(?:hamsterdam|twitter))(?=.\b(?:1[1-9]|2[0-5])\b).*$”

And here is where to plug in the regular expression to Google Search Console:

Google Search Console Custom Regex Filter for Pages

And guess what? It returned no results!

Something was off …

So I gave ChatGPT some examples of the types of URLs I wanted results for:

ChatGPT prompt and response for GSC regex filter

That improved the results somewhat, but some were still missing.

So I asked for a revision again:

Revised ChatGPT prompt and response for GSC regex filter

And that did the trick:

GSC pages tab in performance report

Here is the final regex:

“https:\/\/ethanlazuk.com\/blog\/(?:hamsterdam-part-|twitter-files-part-)(?:1[1-9]|2[0-5])”

So as you can see, ChatGPT can be very useful for creating regular expressions for Google Search Console. But at the same time, it’s important to verify the results, and not be afraid to rely on your own expertise to guide ChatGPT to provide the wanted outcome.

This is also a learning for me in real time because I’ll likely adjust the specificity of my next prompt based on what I just witnessed. 🙂

For more on regex and GSC in general, feel free to check out my article on GSC tips. I also recently saw an advanced guide to using regex in GSC for local search by Andy Simpson that may spark some more cool ideas for prompts.

2. Getting ideas for blog content

I know, I know, using ChatGPT to write blog content, that’s low-effort, right? Yes, it is!

It’s also unlikely to yield the type of helpful, people-first content that Google (or other search engines and answer engines, like Perplexity) is trying to reward.

Keep in mind, AI text generation is fine for search (and thus SEO), but only if the final output is high quality and valuable for the user’s search intent.

Can ChatGPT accomplish that alone? Eh.

But! What ChatGPT can do is give you a ton of ideas for making your blog content more helpful, valuable, high-quality, and ultimately useful.

This is a big topic, so I’ll include lots of subtopics within this section, but here are ways to use ChatGPT for SEO blog content.

2a. Getting inspiration for article content that readers may care about

One tricky aspect of writing so-called “SEO content” is that the goal isn’t to create a piece of content that can just rank in search results and drive traffic. The goal is to get visibility in front of a target audience relevant to that website’s business goals, satisfy their search intent, and inspire them to convert on the website.

For years, word counts and being comprehensive or holistic were spoken of as virtues for SEO content. The idea was that including all the angles and details possible would make the content seem more authoritative and relevant to search engines for the target queries.

But what about the user?

Do they care about all of that information, or do they just want to know what’s relevant to their goals?

This is where ChatGPT can come in handy.

If you have insights about your audience, you can feed those into ChatGPT, along with the target query (or primary question) your blog content is focused on, and then get ideas about what would be relevant to them.

For example, I was creating a content brief for a product comparison article for a client. The focus of the blog article was “Brand A’s product vs. Brand B’s product.”

So I asked ChatGPT the following prompt (edited for anonymity):

“If someone is health conscious, between the ages of 18 and 35, and has an annual income of at least $50,000, what might or should they consider when comparing two brands of (product)?”

For this example, I’ll use “tea” as the product.

In its response, ChatGPT lists out factors like quality of ingredients, nutritional value, flavor, price, ethical considerations, and more:

Comparison in ChatGPT of two types of tea.

With these in mind, I can then inform the writer about which points of comparison might be best to include in the comparison article.

Not only will this allow the writer to naturally incorporate related entities (such as antioxidants or fair trade), but those comparison points are also more likely to be relevant to the target audience than if I just went to a keyword research tool and picked out whatever I could find had the highest volume.

Of course, a subject matter expert who knows the product and audience in depth should also verify whether these points of comparison make sense, but it’s a good start to creating helpful content for the topic.

2b. Competitive SERP analysis of top-ranking results’ content

Let me first state, that I don’t recommend paraphrasing the top results in the SERPs for a target query as a way of creating your own content. Google wants to reward unique content that provides information gain.

In their helpful, people-first content guidance, Google explicitly states that if you are “mainly summarizing what others have to say without adding much value,” that’s likely search engine-first content, which should be avoided.

That said, I’d argue there can be value in analyzing the SERPS to understand the search intent of a query, as well as get inspiration for your own article. (I made this case in my 11x content opinion article.)

Using ChatGPT Pro with the Link Reader plugin — or more recently the main GPT-4 model with Browse with Bing — you can feed some of the top-ranking pages’ URLs into your prompt and ask ChatGPT to assess them in different ways.

If you’re doing topical research, you might ask:

“Please summarize the main points of these articles into bullet points based on their heading structures and contents: (page URLs)”

Or if you’re looking to understand the angle on the topic, you could ask:

“Please summarize how each of these articles addresses the topic of (main query or topic), accounting for their level of detail, style of writing, and other factors relevant to the audience: (page URLs)”

Here’s an example of the first prompt using some of the pages from my own blog. (Keep in mind, the output won’t be that comparable because these pages don’t rank for the same query, but you’ll get the idea):

ChatGPT with Link Reader summarizing several blog articles in bullet points.

Again, the goal isn’t to reuse other website’s content. But by understanding what’s ranking already for a query, you can have more context to inform how you approach your own, unique content.

2c. Creating an inverted pyramid outline for a blog article based on core topics

After you’ve gathered your list of important content sections for a blog, as well as competitive insights from SERP analysis (such as using step 2b above), you can use ChatGPT to organize these into a proposed outline that follows the inverted pyramid structure.

The inverted pyramid structure is a concept from journalism where the most pertinent information goes on top and the more general or less relevant information goes at the end.

This is a valuable way to approach blog content for SEO in many cases because following the inverted pyramid allows the user to find what they need more easily. (In its page experience documentation, Google even mentions that “How easily can visitors navigate to or locate the main content of your pages?” is an element of good page experience and thus helpful content.)

I mean, we’ve all seen those older SEO blogs that start with “What is X?” then “What is the history of X?” followed by “Benefits of X,” when the user just wants that piece of information about the cost of X (or whatever it is) buried at the bottom.

Let ChatGPT help you avoid this.

Here’s a prompt to use:

“Given this list of page sections: (then list the sections), how can I organize them into an article outline that follows the inverted pyramid structure for a reader who cares most about (main query or topic)?”

Here is an example of that where I inserted information related to “ChatGPT for SEO”:

“Given this list of page sections: “what is chatgpt,” “what is seo,” “using chatgpt for article outlines,” “using chatgpt for seo,” how can I organize them into an article outline that follows the inverted pyramid structure for a reader who cares most about “how can chatgpt help my seo”?”

And here is the response from ChatGPT:

ChatGPT prompt and response creating an article outline using the inverted pyramid structure.

As we can see, ChatGPT reordered the sections I provided by putting the key details first and the general definitions last.

This opens up interesting possibilities, like maybe the What is SEO and What is ChatGPT sections aren’t necessary, and the article can simply have the top two sections. Or maybe the final sections could go into an FAQ section at the end.

The output from ChatGPT isn’t a good final form for an article outline, but it does provide inspiration for how to follow the inverted pyramid structure and better satisfy the reader’s needs.

I also recently saw this article with content-related prompts from Screaming Frog. I haven’t tried these yet, but they look pretty good!

3. Creating schema (structured data) for website pages

Schema refers to structured data, a standardized format of machine-readable numbers and values for information. In SEO, uses for structured data include getting eligible rich results on Google but more broadly providing context about entities contained on a page to help search engines like Google with disambiguation and building the knowledge graph, including gaining knowledge panels.

There are different syntaxes of structured data. Google recommends JSON-LD schema, but it recently announced syntax graph merging, where the search engine can merge JSON-LD and Microdata structured data in specific cases.

While many CMSs or SEO plugins will add default schema that you can’t always edit, there are cases where it makes sense to add additional schema to a web page. If you’re using JSON-LD, the recommended format by Google, this will be added to the <head> section of the web page. On my website, I used a plugin to add code to the <head> section. If that’s not an option, you may need the help of a developer, as you want to be careful not to disrupt (i.e., break) your page.

3a. Determining eligible schema types to add

Learning structured data can be like learning a second language. Schema.org has a large vocabulary of schema types. Not all of these are recognized by Google as impacting rich results. However, some structured data advocates say it makes sense to add all qualifying types, as these can only help with disambiguation and knowledge graph building.

If you’re not sure where to start with schema types for a web page, you can turn to ChatGPT.

One way is to simply ask the following prompt: “What type of structured data would make sense for (business type + page type)?”

For example, I asked ChatGPT, “What type of structured data would make sense for an HVAC company’s homepage?”:

ChatGPT prompt for what type of structured data would make sense for an HVAC company's homepage?

In total, ChatGPT listed the following seven types:

  • Local Business Schema
  • Organization Schema
  • Review Schema
  • Breadcrumb Schema
  • Service Schema
  • FAQ Schema
  • Video Schema

Not all of these necessarily apply, but for ones that would, you can then ask for an example of them.

For example, my follow-up prompt was: “Can you give me an example of local business schema for an HVAC company’s homepage?”

As you can see, ChatGPT gives you code in JSON-LD format that you can then copy and fill in the business’s real information:

Example of local business schema in JSON-LD format in ChatGPT.

Now, you may not have all of the information available and have to remove some sections. Or maybe you have other information and are curious if it would qualify.

In those cases, you could instead provide the information about the business to ChatGPT and ask it to generate local business schema.

Either way, make sure to validate the schema before adding it to your website, as well as after publishing it.

While ChatGPT can help with schema validation, you ultimately will want to confirm the schema is validated using the Schema.org validator, or maybe Google’s rich results testing tool.

3b. Creating structured data using page content

In the previous example, we asked ChatGPT to give us ideas for schema types. But what if we know the type we want to use?

One option is to input the data manually and ask ChatGPT to populate it into the structured data.

Another option is to use ChatGPT with the Link Reader plugin to crawl the web page and extract the information itself.

This is how I created the Person structured data for my about page.

When I did this, it took some trial and error. I eventually asked for an example of Person schema (see tip 3a above), filled in the information, and then asked ChatGPT to crawl my About page and add more details:

Using ChatGPT to expand Person schema.

As mentioned in tip 3a above, I also had ChatGPT verify the schema.

A little later, I also incorporated this Person schema into Profile page schema, which Google recently introduced support for.

Bonus: Here’s something interesting about eligible schema. When I looked up my about page on Bing Search in the past, it mentioned my spouse is Dania as a line of text in the snippet. This is directly mentioned in the schema.

Someone has also built a custom GPT for this called Bio To Schema. Feel free to explore that as well!

4. Summarizing a technical article about Google patents

Using ChatGPT with the Link Reader plugin enabled, I’ll often input a technical article about an SEO subject like Google patents and ask for the information to be summarized in bullet points:

ChatGPT prompt and response to summarize an article about Google patents.

If that’s still a little too technical for me to grasp, I’ll ask ChatGPT to revise the response to suit a different reading level. (You may have to play with this so you don’t sacrifice key details for readability):

Revised ChatGPT prompt and response to summarize an article about Google patents.

But that was also in earlier days (actually before Link Reader disappeared and then came back), and since then you now have the ability to give custom instructions to ChatGPT.

These custom instructions can include providing your preferred reading level:

Custom instruction to ChatGPT for responses

It also can infer your goals, like here ChatGPT assumed I wanted to accomplish an SEO task because it knows I work in the field:

This was based on me giving it some of my background information:

Custom instructions for ChatGPT about myself.

Pretty clever.

5. Getting a sense of the general reputation or awareness of a business

Since ChatGPT is trained on what’s known as the Common Crawl, a collection of books, articles, and billions of web pages, and ChatGPT Pro also has the ability to crawl the web with its Browse with Bing setting — this was originally launched in May 2023, then disabled in July, and then recently returned in late September — it can potentially give you additional context into how your brand (or a client’s brand) is perceived online.

In short, if the sentiment around a brand in ChatGPT’s response is negative or factually wrong, that’s a bad sign and something to address.

The original example I found was from the first iteration of Browse with Bing, where I gave ChatGPT the following prompt:

“What does [business name] do?”

Pretty straightforward.

The response was a succinct paragraph that included a superscript “1” that referenced the company’s homepage. This was good and bad, as it shows the company’s own website is recognized as the correct source of information, but that the company also might not have much online reputation beyond that.

Again, this is not a substitute for analyzing a brand’s SERP visibility, knowledge graph representation, or linked/unlinked mentions on the web. This is merely another step in getting a sense of how that business’s brand is perceived online, given that potential customers could be searching for this information using LLMs just as easily as Google Search.

Here’s an example I did for myself.

I asked ChatGPT “What does Ethan Lazuk do for a living?”

At the time, I wasn’t a very established entity or brand, but I was curious how my professional information was represented.

The response was mostly accurate:

ChatGPT Web Browsing prompt and response for what does Ethan Lazuk do

What I find interesting are the chosen sources:

  1. https://ethanlazuk.com/
  2. https://www.linkedin.com/in/ethanlazuk
  3. https://www.tiktok.com/@ethanlazuk
  4. https://ethanlazuk.com/services/

ChatGPT used my website, LinkedIn, and TikTok accounts. Also, three of the four links have jump links that resemble the format of Google’s featured snippet highlighted jump link text or Chrome’s Copy Link to Highlight feature, e.g., #:~:text=. (Note: I had to remove this portion from the naked URLs above to prevent mobile usability issues, but if you visit those links, you’ll see the portion with the jump link included.)

Curious if you can influence LLM outputs?

Later on, I created an article to explain how I think about SEO. This later got incorporated into both ChatGPT and Copilot (Bing Chat):

An excerpt from my How Ethan Lazuk Thinks About SEO article showing ChatGPT's response.

I’ve also found ways to influence Google’s SGE, which I’ll discuss more later.

6. Getting reporting insights from Google Analytics 4 (GA4) data

I hear from other people that one of the ways they use LLMs like ChatGPT or Bard is to get a first draft of writing and the refine it from there.

That’s not my style.

I love writing, and I wouldn’t ever base my writing on a generative AI output. The furthest I’ll go is having Grammarly find spelling errors or typos in my writing.

That said, when it comes to reporting on website performance data from Google Analytics 4, I can use some help. Typically, I can report SEO insights related to organic search as a medium. However, when it comes to cross-channel insights involving paid, direct, social, email, or other channels of traffic, I need a hand.

One solution is to simply take a screenshot of your GA4 data, upload that to ChatGPT Pro (GPT-4), which is now equipped with DALL-E, and ask it to summarize insights.

Here’s a prompt you can use:

Based on the GA4 data in the screenshot, what insights can you tell me?

That’s it.

Here’s an example where I asked ChatGPT to analyze the Path Exploration (Explore Report) for my website so I could better understand how users navigate between pages:

ChatGPT insights based on a GA4 Path Exploration report for EthanLazuk.com.

The insights were pretty remarkable. In client contexts, I’ve gotten even more specific results by inputting screenshots of GA4 Acquisition or Engagement reports for traffic or conversions.

7. Generating custom featured images for blog articles using DALL-E

It’s incredibly easy to generate custom AI-generated images using ChatGPT Pro today thanks to DALL-E.

Simply give a prompt describing the image. I also like to include suggested dimensions.

The tricky part is refining the parameters of the image to get it the way you want.

I’ve found that AI-generated images tend to be on the extreme side with regard to detail and fantastical elements.

Here’s an example:

Example of a busy image created in ChatGPT.

So I typically ask ChatGPT to repeat the previous image while making whatever changes I want, and then I suggest that it “simplify” the image.

The word “simplify” is hugely important for generating the types of images I like to add to my blog articles.

Sometimes you also have to download one of the images and re-upload it in the prompt to ask for changes, so it stays similar in theme.

What’s nice about custom blog article images is it’s a way to set your content apart from everyone else’s, such as with image thumbnails that appear in Google’s search results, in SGE answers, or even in Discover feeds.

Custom images can also lead to clicks to your article from Image search. That’s less likely to happen if you’re using copycat images.

As for inspiration, one method I like to use is to tell ChatGPT a band I like as part of the artistic direction:

Here’s the original featured image I had for this article, taken from Pink Floyd artwork:

Pink Floyd Welcome to the Machine featured image in a previous version of this article.

Here’s the latest custom featured image I came up with:

Welcome to the Machine themed AI artwork from DALL-E.

To achieve that, I used several prompts but suggested the artwork be in the spirit of Pink Floyd’s Welcome to the Machine song. 😉

I’d also suggest dropping these ChatGPT (DALL-E) images into TinyPNG or another image compression software before adding them to your website. This will help the image file sizes from weighing down the page speed and impacting your webpage’s performance and user experience.

8. Buyer’s journey clustering of PAA questions

Google People also asked questions are a popular resource for understanding audience questions and mapping buyer’s journeys. (Relatedly, I’ve also written about how Perplexity AI’s Related questions can add additional context to PAA.)

One tool that can assist with PAA question gathering is AlsoAsked.

To help build topic clusters along buyer’s journeys and understand audiences’ search intents, I’ll export PAA questions from AlsoAsked using a deep search:

AlsoAsked deep search of Google PAA questions for beach kite.

Next, I’ll upload those to ChatGPT, and then ask the AI to map the questions to their respective stages of the buyer’s journey.

Here’s an edited prompt I’ve used for a client:

I have a company that’s a [description]. I did some research into what people want to know about this topic. I’ve attached that list of questions and topics. I want to create articles and guides for my website that will help answer these questions for people. Can you please group the questions together into logical topical or semantically related groups. Once you’ve done that, please give me an idea of how those topics or groups fit along a buyer’s journey someone would take from awareness [examples] to decision [examples].

Here’s an example of how this would look for the hypothetical topic of buying a kite for the beach:

ChatGPT analysis of PAA questions on the buyer's journey.

Next, I’ll use those topics to inform my content strategy and the order in which I create topics within a cluster to complete a buyer’s journey.

9. Visualizing GSC data and summarizing for stakeholders

One challenge I run into as an SEO is translating findings from monthly reporting into visualizations or summaries that a high-level stakeholder, like a CEO, would care about.

In this example, I asked ChatGPT to visualize the CTR for branded vs. non-branded queries from Google Search Console (based on instructions for what constituted a branded query):

ChatGPT visualization of branded vs. non-branded CTR from GSC data.

Then I asked for it to summarize these findings in a way that a CEO would care about:

ChatGPT explanation of CTR performance data in a way a CEO would care about.

This conversation was part of a longer thread that looked at top queries, CTR for topical clusters, and other relevant data, which I’ve kept concealed here for privacy reasons.

However, the capabilities can be adapted to your needs for reporting.

Relatedly, I’ve also written about how to do this branded vs. non-branded comparison of queries in a GSC for SEO guide.

10. Reformatting video transcripts for content work

This is a fairly straightforward tip but I included it because I think it also highlights a key point to remember, which is selecting the right AI model for the job.

I had to summarize some YouTube videos using the auto-generated transcripts, which are messily formatted after copy-pasting with time codes, annotations, and no grammatical formatting.

By pasting an unformatted video transcript into ChatGPT and asking it to format the transcript verbatim into a single line while removing unnecessary information and ensuring correct grammar, you then have workable content that you can repurpose for direct quotes or use for content ideation and other purposes.

Here’s an example:

ChatGPT prompt to clean up video transcript.

If the text is long, you can also ask for it to be divided into logical paragraphs, depending on your needs:

ChatGPT dividing formatted video transcript into paragraphs.

But what’s interesting is I defaulted to using GPT-4 for this project when I did it for a client, but when I hit my quota of prompts, I switched to GPT-3.5. It was actually much faster and just as accurate.

So if you’re doing simple text reformatting, like for video transcripts, consider testing if GPT-3.5 can serve your needs before you try GPT-4. It may just save you time.

SEO-related custom GPTs

Here’s a list of custom GPTs for SEO and related tasks. At the moment, these are all free.

As a disclaimer, I don’t really use GPTs that often myself. (I usually use prompts instead.) However, I’ve experimented with these GPTs, and they come from sources I trust.

I’ll add more GPTs to the list as I discover them!

Claude AI uses for SEO

Whereas my uses of ChatGPT tend to be more in the data analysis realm, how I’ve used Claude has been more for content revisions:

1. Revising existing web page content to improve readability and clarity

A while back, I was working on optimizing a comprehensive guide page for a medical product.

Over the years, the information on the page has been expanded and revised by several different authors. As a result, while the facts themselves were authoritative and the guide comprehensive, it read like a mishmash of thoughts.

For a reader, it might have been challenging to read one section about one aspect of the product and then jump to another section only to encounter a completely different tone of voice and vocabulary.

And for the business, their brand identity was lost.

To make the page easier to read, I went section by section and inputted this prompt into Claude:

“How might you slightly improve this text:”

And then pasted in the first section.

Claude rewrote the section and then provided a list of its improvements along with a rationale.

Based on this, I asked Claude the following prompt:

“Please repeat this exercise for this text.”

That’s it.

Over and over it went, section by section.

The reason I did it this way was due to the sensitive nature of the topic. I didn’t want to insert a large amount of text and risk important details being misrepresented to the end user.

Nor did I ask Claude to add any information. For instances where I felt more information or context was warranted, I used other web content or resources from the same business to ensure consistent information, then added them to the text myself, which I had Claude revise.

In the end, the text came out with slight improvements and greater consistency.

As a personal rule, I’ll never use AI to revise content for my own website. I want this site’s content to represent my voice and style. That said, I don’t have the cleanest writing style 😉 — partly the reason I switched focuses from content creation to SEO strategy work years ago.

But for the sake of demonstration, I’ll take some content from one of my articles and give it to both Claude and ChatGPT so you can see the difference.

Here is the original text, taken from my case study of GA4 vs. Shopify revenue for organic search:

“I did find it interesting how the range of revenue attributed to SEO efforts for the first site could be roughly anywhere from 30% to 200% or more higher based on using any of Shopify’s sales attributed to marketing reports versus GA4’s Ecommerce purchases, while for the second site it could be as much as 25% higher or 40% lower.“

– Excerpt from my GA4 vs. Shopify organic search revenue article.

Here is how Claude revised the above paragraph:

Claude AI prompt and response to revise text

The actual text Claude revised was as follows:

“I found it interesting how much the revenue attributed to SEO efforts differed between the two reporting tools for Site 1 and Site 2. For Site 1, Shopify’s sales reports showed SEO driving 30-200% more revenue compared to GA4’s ecommerce purchase reports. For Site 2, Shopify credited SEO with up to 25% higher revenue, while GA4 showed SEO contributing up to 40% lower revenue. The significant variance in SEO revenue attribution between the two reporting tools is noteworthy for both sites. Further analysis is needed to determine which report more accurately assigns revenue credit to SEO activities.”

– Claude 2 response (see above image)

Personally, I don’t find the text that interesting, and some liberties were definitely taken with the facts. But I do find the suggestions helpful.

In this case, I might play with the prompt to lessen the degree of changes to the original text.

For comparison, here’s how ChatGPT would revise the text with the same prompt:

ChatGPT prompt and response to revise text

Here’s the text in the way that ChatGPT revised it:

“It was intriguing to see that, based on Shopify’s marketing sales reports versus GA4’s Ecommerce purchases, the revenue attributed to SEO efforts for the first site could vary significantly, ranging from an increase of 30% to over 200%. For the second site, this could result in a 25% increase or a decrease of up to 40%.”

– ChatGPT response (see above image)

In my opinion, neither Claude’s nor ChatGPT’s text blew me away. The ChatGPT text is a bit more accurate, while Claude’s text is more interesting to read.

I also think this illustrates why you won’t find any suggestions for using AI strictly to generate content for SEO purposes in this article.

While AI is helpful for summarizing or revising text or maybe creating an introduction, conclusion, or suggested outline, I’d never rely on it to write context without the heavy involvement of a human author and expert reviewer.

2. Creating examples of LocalBusiness schema (or any structured data)

We’ve talked about creating schema with ChatGPT earlier, but Claude AI can do the same.

In this example, I wanted to generate localBusiness schema for a website, only instead of plugging in the details and asking the LLM to create the schema, I simply asked for an example of what the Schema might look like:

Claude AI generating an example of localBusiness schema.

Then I filled in the details using the business’s information, verified it on Schema.org, and voila.

Easy peasy.

3. Generating content suggestions from technical documents

Keyword research can offer insights into how customers search for topics. But what happens when there’s no obvious keyword to target? I wrote a blog post about this, but I’ve also been using Claude to help out in such situations.

I was recently working with a client that is the only one in their field. Keyword research was pretty generic, but we need to create some content topics throughout the buyer’s journey to create product awareness.

What this client did have was a treasure trove of technical documents.

By plugging these documents as PDFs into Claude and asking not only for a summary but also suggestions for who might find this information helpful and why, we could find blog topics that our personas might find useful during the purchase funnel.

You can take a two prompt process.

First, ask Claude to explain who would find the document useful. This helps give insight into the personas.

Next, ask for how the document’s information could inform blog post topics that align with those use cases.

Here’s an example using a podcast transcript from Google’s Search Off the Record podcast:

Which audiences might find this information useful Claude prompt.
Blog topic ideas prompt to Claude AI.

In a post-helpful content system era where Google is trying to surface more original content that users find useful, this can be a great way to get more unique yet relevant content ideas without relying on keyword research.

Gemini (Bard) uses for SEO

I didn’t use Bard that often at first, I’ll admit. I started using it more after the Gemini model announcement in early December 2023, and then I got a free trial starting in February for Gemini Advanced, which also incorporates Gemini into Google Workspace, like Google Docs, Sheets, Meet, and Drive.

Something nice about Gemini (Bard) is you can return to your work, even months later, and continue a conversation.

For some of these examples, started months earlier, I just picked up where I left off.

Something else nice about Gemini is you can click the Google icon on the bottom of the chat to verify (double check) the reply based on Search results and even visit that source:

Bard checking an answer using Google search.

Pretty cool.

Here are some of the ways I’ve used Gemini for SEO:

If the screenshot is white, that’s a free version of Bard. If it’s black or from another Google application, that’s Gemini Advanced (post-February 2024).

1. Determining a Google Business Profile (GBP) primary business category

The business categories of a GBP help tell customers about the main products or services of that business. Categories can change over time, and they’re also factored into Google’s local search rankings (organic position in local packs) of that business.

I wouldn’t let Bard just pick a category for me, especially since it can hallucinate them. For example, these suggestions weren’t too helpful, as none are actual categories:

Asking Google Bard to select a primary business category for a vegan bakery.

That said, it’s hard to find a list of categories. You have to find a website that’s created one, like here.

That said, Bard can give you inspiration for applicable categories.

Hey, it’s a start.

You can also try inputting some of the real categories and seeing which Bard suggests is the best choice:

Follow-up question in Google Bard about the best category for a vegan bakery.

Here, that seemed to work.

2. Choosing locations to target in a local SEO strategy

Oftentimes in a local SEO strategy, you’ll have to create location pages or otherwise come up with markets to target for services.

Bard (now Gemini) can help you with this work.

In this example, I had a list of eligible markets for a client, so I asked Bard to organize them by population:

Asking Google Bard to order cities based on population.

You can even refine the results by persona, such as asking which area’s demographics most closely align with your target audience:

Asking Bard a follow-up question about which city is the best fit for an audience.

Of course, you should double-check your work, but it can get you started.

3. Creating chart templates in Google Sheets

I had some AI capabilities in my Google Workspace already due to Google Search Labs, but once I signed up for Gemini Advanced, the capabilities expanded automatically.

A prospective SEO client asked me to put together a proposal outline that listed different tactics along with their priority level and other details like estimated hours or cost.

In the Gemini assistant chat in Google Sheets, I asked for a chart that contained the columns I needed. It generated such a chart in a format that I could then plug in my own data.

Google Gemini in Google Sheets creating a chart for an SEO proposal.

In addition, you could modify or style the chart however you wanted. This is a good example of efficiency, where Gemini is creating the outline to fill in.

4. Making custom formulas for Google Sheets calculations

Within the chart above, I also needed to enable calculations of cost estimates based on hours. Not a difficult formula, but still one I would have had to look up.

Instead, I just asked Gemini:

Gemini in Google Sheets creating a formula for a dollar calculation.

I then plugged that formula into the cell and dragged it to all applicable cells.

Again, it’s not a dramatic task, but it did save me probably 5 minutes, which I could then use for other business.

5. Clustering PAA questions by buyer’s journey stage in Sheets

Earlier in the ChatGPT section, I showed how I’ve used it to cluster Google People also ask questions exported from AlsoAsked into buyer’s journey phases.

However, now with Gemini in Sheets, we can use it for a similar purpose.

Here’s the prompt I gave:

Prompt in Gemini in Google Sheets for PAA questions along a buyer's journey.

And here is the output:

Gemini clustering PAA questions by buyer's journey phase in Google Sheets.

This allows us to more easily edit the information within our original document, versus manually copying or trying to export it from ChatGPT.

Then we can also cross-reference the two data sources, giving us an even richer depth of information from which to build a content strategy with topic clusters along a buyer’s journey.

Now, these uses of Gemini in Google Sheets (and Workspaces generally) have only become available since February 2024, so I anticipate this section of the guide will grow over time.

Again, it’s not a dramatic task, but it did save me probably 5 minutes, which I could then use for other business.

6. Summarizing YouTube videos

Video content has been relevant to SEO for years due to video carousels in SERPs, not to mention referral traffic or just brand awareness generated from YouTube.

However, video content is appearing in more places on Google these days, including Discover feeds, People also view feeds, as well as in Perspectives.

During research for content topics, audience personas, or other key information for SEO strategies, YouTube videos from your brand or a competitor can also be gold mines of information.

However, it’s much easier to work with text summaries of a video’s main contents than to reference video clips.

That’s where I find Gemini helpful. Because it’s powered by Google and is multimodal, it can innately process YouTube videos.

You can search for videos in Gemini, but I like to find videos of interest through Search and then drop them into Gemini for a summary:

Gemini summary of a YouTube video.

The resulting insights can be helpful for creating blog content ideas or gathering webpage details or direct quotes for testimonials or reviews.

I recently worked with a client that had little website content to work with but 10 years of YouTube videos. Gemini made it easy to process all of that information in a text format.

7. Creating analogies to understand complex topics

Something I noticed that Gemini started doing on its own during discussions was putting its answers into analogies. This was unprompted, by the way.

So, I started asking for analogies directly, particularly when dealing with complex or technical topics.

Sometimes the analogies are a little basic, but you can make them more detailed with re-prompting. And often, I find analogies are among the most helpful ways to understand a topic.

Here’s an example where I asked Gemini to explain how different AI concepts relate to one another. It created an analogy of a house.

Gemini prompt and answer with analogies for understanding AI concepts.

From there, I continued the conversation across many other follow-up questions, each time going deeper into AI concepts. It’s a great way to explore a topic.

Just remember to fact check the information yourself.

Gemini 1.5 Pro (Google AI Studio) uses for SEO

I originally lumped in Gemini from Google AI Studio with the above section, but there are some differences between them.

Gemini 1.5 Flash and Gemini 1.5 Pro with a 2 million token context window are alternative models available in AI Studio, with different capabilities and results.

Here are ways I’ve used Gemini in Google AI Studio for SEO-related work, largely for my own projects and website:

1. Inserting larger documents or multiple files in prompts

I received access to Gemini 1.5 Pro in Google AI Studio on March 6th, 2024. I’ve been experimenting with it since, mostly the default Chat prompt experience.

Gemini 1.5 Pro Model settings in Google AI Studio.

At first, I was using it for the same uses as discussed throughout this guide, but one opportunity it presents is being able to handle larger documents.

For example, I created and uploaded a PDF that was 30,922 tokens to Gemini 1.5 and had it summarize it.

To upload the same information to ChatGPT, I had to break the document into separate PDFs.

This is a limitation of the token size of the context window.

That was before, though.

Today, Gemini Advanced has a 1 million token context window.

However, Gemini 1.5 Pro in Google AI Studio has a 2 million token context window, and you can upload considerably more documents to it.

I recently uploaded nearly 1 million tokens worth of podcast transcripts, for example, or 75 PDF documents in total. I wrote about some learnings from that experiment in another post.

If you’re working with larger file sizes or multiple files, whether text, video, or otherwise, Gemini in Google AI Studio can be handy, although it’s not as user-friendly as Gemini Advanced.

2. Breaking full videos into clips or shorts

If you have longer videos that you’d like to break into shorter clips for TikTok, YouTube Shorts, or other redistribution, Gemini can be the answer.

Given the size of the context window, it can handle longer-form video files.

Then you can prompt the model to identify time stamps of certain lengths that would be good for making short clips.

Here’s an example where I uploaded a 25 minute video I recently made about SEO tips for musicians and had Gemini identify good sections for 1-minute clips for YouTube Shorts:

Gemini identifying timestamps for short clips from a full length video.

You could also use these for timestamps of key moments or chapters, which leads to our next tip.

3. Outlines of videos for key moments or related blog content

If you have longer videos, a transcript can be helpful, but wouldn’t readers likely prefer to have an accompanying summary or related blog post?

I recently uploaded a full-length video I shot with SEO tips and asked Gemini to identify time stamps.

However, the outline it provided gave me a great idea for an accompanying blog post or enhanced transcript:

Gemini outline of thematic chapters from a video.

Given how multimodal search user journeys are becoming, Gemini provides a great way to create multiple formats or lengths of content from a single original file, in this case, a video.

Combining multiple chatbot answers for consensus

What’s the one thing that every generative-AI answer says at the bottom? It’s a disclaimer to check if its factual accurate.

One way you can do this is to use one chatbot for the first answer (like ChatGPT) and then put that into the prompt of another chatbot (like Gemini) and ask its opinion.

Another use for multiple chatbots is to get more context and consensus. Same approach here. Take one chatbot’s answer and input it to the prompt of another chatbot and ask its opinion.

Here are some use cases for using multiple AI chatbots to get consensus for SEO work:

1. Topic research recommendations for content

I have a side project called Hamsterdam Research, where we look at a recent AI research paper to understand its concepts and evaluate their hypothetical implications for the future of SEO and search engines.

How do I choose topics, though?

Well, one way is I visit the homepage of a list of AI research papers, like the Google Research publications page.

Then I copy the list of papers and input it into both ChatGPT and Gemini and ask each of them which one is most relevant to SEO and why.

Based on this, I compare the answers and ask follow ups to land on the best and most relevant topic.

Let’s compare two initial answers from ChatGPT and Gemini that it gave me for my most recent article.

First I put the list into Gemini:

Gemini answer to which Google Research paper is most relevant to SEO.

We can see it recommended a paper about active speaker detection.

When I asked ChatGPT next, it gave me some different answers:

ChatGPT answer to which Google Research paper is most relevant to SEO.

But then when I prompted it based on Gemini’s top answer, it agreed it’d be relevant:

ChatGPT follow up answer to which Google Research paper is most relevant to SEO.

So based on that collective feedback, I decided to write about ASD (assisted speaker detection).

You could do this for your own business and blog. For example, take a list of content topics from your content calendar, input them into multiple chatbots, and ask which one is most relevant to your audience or your business goals, and prioritize accordingly.

2. Creating a client questionnaire and exporting it

As an independent SEO consultant, I work with a lot of partners and onboard clients. This involves providing them with a questionnaire to learn their goals and other relevant information before starting an audit or building a strategy.

I recently had to create a questionnaire to give to a partner. So I used ChatGPT and Gemini together to build the most holistic set of questions based on what I needed.

I first told ChatGPT what I wanted:

ChatGPT answer about an SEO questionnaire.

Next, I took that response and fed it to Gemini:

Gemini answer about an SEO questionnaire.

What’s great about using Gemini last here, is I can now export that response directly to a Google Doc, where I can edit it based on my specifications:

Gemini export to Google Docs.

You can also use Gemini in Google Workspace to further refine it:

Gemini in Google Docs.

This is a great way to get multiple sources of context rather than relying on a single LLM response, as well as save time on an initial draft of a client deliverable.

3. Fact checking key takeaways

LLMs confront both epistemic uncertainty (lack of knowledge) and aleatoric uncertainty (randomness).

So when you ask a question, especially one that may not have a known answer, how can you be sure you’re getting good information?

One technique I’ve used is to take the answer from one chatbot (like Gemini Advanced) and put it into another chatbot (like ChatGPT) to ask if it draws the same conclusions.

I recently did this for a SER story about an update to Google’s image SEO documentation.

You could also bring others into the mix, like Claude. If you see consensus, that’s a good sign. If the LLMs draw different conclusions, you might want to dig in a little more.

Tips for prompting AI chatbots (LLMs)

Google Workspace created a “Prompting guide 101” that has a lot of helpful information that can apply to all AI chatbot conversations (Gemini, ChatGPT, Claude, etc.), in my opinion.

It’s a 45-page PDF, so naturally I uploaded it to Google AI Studio and asked Gemini 1.5 Pro to summarize the key points:

Gemini 1.5 Pro summary of effective prompting tips.

From Gemini’s answer, a few takeaways are:

  • Speak conversationally
  • Be specific, clear, and concise
  • Follow the four pillars: persona, task, context, and format
  • Aim for around 21 words per prompt
  • Review and refine
  • Break complex tasks into steps

Here’s the full prompt guide, if you wish to review or download it:

gemini-for-google-workspace-prompting-guide-101Download

Prompting the LLM with an audience, context, format, and a goal is very helpful. It’s also helpful to speak to AI chatbots more like a human than a search engine, while always remembering their limited context, pun intended.

Outro

I’ll continue to update this article with more use cases of LLMs for SEO. I have a lot more that need to get added, including (if I recall correctly):

  • Putting keywords into topical clusters for internal linking and content strategy (pillar-cluster organization)
  • Combining SEO data from Screaming Frog, GSC, and Semrush into usable spreadsheet documents for reporting
  • Extracting data from spreadsheets and transferring it to formatted Word documents for uploading to website pages
  • Translating GSC performance data into new graphical representations
  • Getting Python coding assistance in Google Colab
  • Uploading a screenshot of a CSS issue with a website and asking for coding help
  • And lots of others + new ones I’ve used since

Feel free to leave a comment with feedback or questions, or contact me (I’m always happy to chat, learn, or help out), and check out related posts below.

Until the next update, enjoy the vibes:

Thanks for reading. Happy optimizing! 🙂

SEO Strategist & Consultant

Ethan Lazuk

Related posts

Editorial history:

Created by Ethan Lazuk on:

Last updated:

Need a hand with a brand audit or marketing strategy?

I’m an independent brand strategist and marketing consultant. Learn about my services or contact me for more information!

Leave a Reply

Discover more from Ethan Lazuk

Subscribe now to keep reading and get access to the full archive.

Continue reading

GDPR Cookie Consent with Real Cookie Banner