How To Do GEO Keyword Research And Track Your AI Visibility

By Corey Frankosky · Last Updated August 17, 2026

AI answers now sit above the blue links for a lot of searches. Instead of ten results with full descriptions pointing at your website, you get an answer with a handful of quick links and sources underneath it. If you are not one of those sources, good luck getting found. GEO keyword research is how you figure out which prompts people are actually asking, whether you show up in the answers, and what to build so you get cited. This guide walks through the research process, the tracking setup, and my honest take on whether the paid AI visibility tools are worth it yet.

1What GEO keyword research actually means

GEO stands for generative engine optimization. People also call it AI keyword research, AI search optimization, or a few other names. Nobody knows what it will be called in three years. Right now I call it GEO keyword research.

One thing worth clearing up. A lot of articles use "GEO keywords" to mean geographic keywords, as in service plus location terms like "HVAC repair Dallas." That is a different thing entirely. This article is about generative engine optimization, which is about getting cited inside AI answers. There is a section further down on local service and city prompts, because those overlap in practice, but the core topic here is AI citations.

Here is what changed. When somebody searches "what's the best way to learn Google Ads" and clicks into AI Mode, Google builds an answer and shows you exactly what it sourced. Every video it found. Every page it went through. If your site or channel is not on that source list, you are invisible for that question even if you rank on page one below the answer.

The data backs this up. Pew Research analyzed 68,879 Google searches from March 2025 and found that users clicked a traditional search result 8 percent of the time when an AI summary appeared, compared to 15 percent when it did not. Only 1 percent clicked a link inside the summary itself.

So the goal shifts. You still want to rank as high as possible on Google. That has not changed. But now you also want to be recommended. Those are two different jobs.

Question to Answer:

When you search your most important topic in Google AI Mode, is your website or channel listed in the sources?

2Watch the full tutorial

I walk through real examples in this video, including live AI Mode searches, the prompt research tools I use, and a look at what the paid AI visibility trackers actually cost.

GEO Keyword Research: How To Get Cited In AI Search Results

Question to Answer:

What are the three or four topics you most want to be the recommended answer for?

3You are researching prompts, not keywords

This is the mental shift. Traditional keyword research gives you a term and a search volume. GEO research gives you a question and a follow-up.

People do not stop at the first prompt. They ask "what's the best way to learn Google Ads," read the answer, then ask "what if I want to learn on YouTube." That follow-up is a completely different competitive landscape than the first question. When I ran that exact sequence, Google Skillshop came up first on the initial prompt, which makes sense because that genuinely is where you should start. On the YouTube follow-up, I came up number one.

Where the AI pulls from is broader than a standard search result. It is looking at YouTube channels, Reddit threads, Quora, LinkedIn, and individual websites, especially the established site in a given industry. That means the sites competing with you for AI citations are often not the same sites competing with you in organic rankings.

It also means the goal is not "create a page targeting this prompt." It is "make sure this question is answered well somewhere in content I already publish."

Question to Answer:

What follow-up question would somebody ask after your main topic, and are you covered for that one too?

4How to find the prompts people actually ask

AnswerThePublic is one of my favorite tools for this. Not because it gives you keywords, but because it gives you real questions. Every one of those questions is a potential video, a short, a tutorial, or a section inside an article you are already writing.

The faster version is to just ask an AI directly. Go into ChatGPT and say "give me 25 questions people consistently ask about Google Ads." You will get something close to what a prompt research tool gives you, for free, in about ten seconds. Swap in your own topic.

Other places to pull real questions from:

• Google Search Console. Your queries report is full of the actual phrasing people use. This is real data, not estimates.
• AlsoAsked. Good for mapping how one question branches into related ones.
• Google Keyword Planner. Still useful for finding the commercial and transactional variants.
• Your own comment section and support inbox. These are the real questions your audience asks, not a guess at them.
• Semrush and Ahrefs. Useful for competitor gap analysis and for spotting which queries trigger an AI Overview.

Here is the part people get wrong. You do not need to create a new page for every prompt on that list. If the prompt is "what are the best tools to analyze pay-per-click ad performance" and you already have a video and a page about PPC tools, you are covered. Do not build a fresh article for a question you already answer. Just make sure the answer is actually in there, clearly, in a form that can be lifted and quoted.

What you are looking for is the gaps. If the prompt is "how do I set up conversion tracking for an e-commerce website" and you have nothing on that, now you have a real content idea: ecommerce conversion tracking for Google Ads and Meta Ads.

Question to Answer:

Are you collecting the real questions your audience asks, or guessing at them?

5Map prompts to content you already have

Build a simple spreadsheet. One row per prompt, with columns for the prompt itself, the article on your site that answers it, and the video on your channel that answers it.

Say you are covering CRMs. You have an article about the best CRMs and a video about the best CRMs for 2026. Maybe you also have videos on best CRMs for local service businesses and best CRM for accountants, because different tools work for different industries. Now when somebody prompts "what is the best CRM for a small agency," you have an article and a video that both speak to it. That gives you a real chance of showing up.

The pattern that works is one topic covered in multiple formats, all interlinked. An article on the site, a video on the channel, the video embedded in the article, the article linked from the video description. That is not a new SEO idea. It just matters more now, because AI systems stitch answers together from multiple sources and multiple formats.

If you are the product, the math is different. Say you are a CRM built for small agencies. You can publish all the content you want, but a big part of how you get recommended is somebody on Reddit saying "we've been using this and it's been going great." I do not think one Reddit comment decides anything. But more people mentioning you across more places absolutely helps, and that is not something you can write your way into. Part of SEO is just having a good product or service. That part is not marketing.

Question to Answer:

For your top 10 prompts, do you have both an article and a video, and are they linked to each other?

6Service and city prompts for local businesses

If you run a local service business, your prompts have a location in them, and they break into two types. Explicit terms name the place, like "roof repair Houston." Implied terms use context, like "roof repair near me." Both matter, and you should track both.

Start by listing every service that brings in revenue, then add location modifiers: city, ZIP code, neighborhood, and near me. Assign one primary term to one page so your site, your ad group, and your AI prompt list all point at the same thing.

Then turn each of those into a prompt set. The same term produces different prompts at different stages of the decision, and you want coverage across all of them:

One thing to keep in mind from my own testing: most of my AI citations come from informational searches, not transactional ones. I use AI tools constantly and I am almost never trying to buy something in them. So if you are a local business, the awareness and shortlist prompts are probably where you have the most realistic shot.

Question to Answer:

Have you run your top service and city prompt in an AI tool, and did a competitor come up instead of you?

7Comparison and best-of prompts, done honestly

Buyers compare inside AI tools, so comparison and best-of prompts matter. But this is where people go overboard, and Google's updates have already hit a lot of it.

Do not go build 50 comparison pages so you can be cited for all of them. Ask instead: what are the three or four things we are really trying to be known for, and do we have the right comparisons for those? For me that is "learn Google Ads" and "Google Ads course." I can build comparisons around those. I am not going to write "Surfside PPC vs. [competitor]" pages for every agency in the space.

And I would not write "we're much better than [competitor] because of XYZ" for a niche I do not actually dominate. First, it is probably not true. Second, Google is not going to treat it as helpful content, and helpful content is what gets raised to the top of AI visibility long term.

Question to Answer:

Do you have a best-of or comparison page for your two or three core topics, and would a stranger consider it fair?

8How to track AI visibility without paying for a tool

You can get most of the way there for free with two things: your analytics and a spreadsheet.

Check your GA4 referral traffic. Google Analytics 4 shows referral traffic from ChatGPT, Perplexity, and Gemini. Go to your traffic acquisition and user acquisition reports and look at the source. ChatGPT referrals come through with a UTM source attached, so they show up cleanly. I get some of this for Surfside PPC, and it tells you two useful things: how much traffic AI tools are actually sending you, and which pages they are sending it to.

Run manual prompt checks. Take your top prompts and run them in ChatGPT, Google AI Mode and AI Overviews, Bing Copilot, and Perplexity. Do it logged out so personalization does not skew what you see. Results vary a lot between platforms, which is why checking one is not enough.

Log each check in a sheet:

Also log which third-party source the AI cited. If it pulled from a directory, a Reddit thread, or a competitor's roundup instead of your site, that tells you exactly where to put your effort next. Track mention rate, citation rate, and competitor citations as your three numbers. If you run 20 prompts and get cited in 4, that is a 20 percent citation rate and that is your baseline.

Check one technical thing before any of this. Open your robots.txt file and make sure you are not blocking GPTBot, ClaudeBot, PerplexityBot, or Google-Extended. If those crawlers are blocked, you have taken yourself out of the running before you started.

Question to Answer:

Is your robots.txt allowing AI crawlers, and do you know how much referral traffic you already get from ChatGPT?

9AI visibility tracking tools and whether they are worth it

There is a whole category of AI visibility trackers now. They monitor a set of prompts across ChatGPT, Perplexity, Gemini, AI Overviews, and Copilot, then report on whether you were mentioned and who got cited instead. Some of them bundle in SEO audits, content generation, or an agent that tries to improve your rankings. Semrush and a few of the established keyword tools have added AI Overview tracking to their existing plans too.

Here is my honest take. Entry pricing tends to run around $29 a month for something like 15 tracked prompts. For me, that is not worth it. I am not tracking 100 prompts to see how I am doing, and I can run a quick search myself every day for free.

I have not done a ton of formal AI visibility tracking for Surfside PPC or for clients yet, and I want to be upfront about that. It is a new enough category that I do not know how well these tools actually work. What I do is check my Google Analytics reports to see what traffic I am getting and which pages it lands on, and run manual checks on the prompts that matter.

If you want to try one, the question to ask is simple: is there a cheap way to track the 20 to 40 searches we actually care about? Not 500. The main things. Start there and see whether the reporting changes any decision you make. If it does not, the spreadsheet was enough.

Question to Answer:

What would you do differently if a tool told you your citation rate, and is that worth the monthly cost?

10How to actually earn the citation

The tactical part is straightforward, and it will sound familiar. Answer clearly. Give a direct answer the model can lift and quote. Use real headings, short definitions, lists, tables, sourced claims, and original data where you have it. SEO is basically the same as it ever was.

The part that actually moves the needle is volume and coverage across formats. Your goal is to reach everybody interested in your topic, in whatever format they prefer to learn in. For me that means the YouTube channel, the blog, the podcast, and more social content than I currently produce.

Think about what it looks like when it works. You have 100 videos across 100 topics, 100 blog articles, and they are all interlinked, and you have the best information available on those topics. At that point you are not optimizing for one prompt. A rising tide lifts all boats, and AI systems have a lot of surface area to find you on.

I have had people leave comments saying "Gemini recommended this video" or "ChatGPT sent me here." That is the whole thing working. Somebody asked a question, an AI built a shortlist, and I was on it.

Question to Answer:

Are you covering your core topics in more than one format, and is that content linked together?