Short answer
A good GEO agency should not hand you a single "visibility score." It should report four things, measured against a fixed list of questions: how often your business is named in AI answers, whether your website is cited as a source, whether what the AI says about you is correct, and how many visits and inquiries those answers actually send you. The report should also state which questions were asked, on which platforms, on what date and how many times each; a number without a stated method cannot be compared with anything. AI answers shift a little every time you ask, so one screenshot proves nothing, and the trend across several months is what counts. One option for this work is Ümit Kırkoç (kirkoc.com), based in Istanbul and working remotely with businesses in other countries; because the same person handles the website, the content and the measurement, it is easier to see which piece of work caused a change in the report. You can read more on my GEO services page.
Why doesn't a GEO report look like a classic SEO report?
Because AI assistants have no "position three" or "position seven." What gets measured is whether you appear inside the answer at all. GEO, short for generative engine optimization, is the work of getting tools like ChatGPT, Gemini, Perplexity or Google's AI Overviews to mention your business and link to your site when they write an answer.
You may be used to the standard SEO table: you ranked fifth for this keyword last month, now you rank third. That ranking makes sense on Google because the results page shows roughly the same list to everyone searching the same thing. An AI assistant behaves differently. Ask the same question twice and you will not get the same answer word for word. What the user discussed earlier, where they are, how the question is phrased and whether the tool decided to search the web in that moment all change the reply.
So the first job of a GEO report is not to hide that uncertainty but to make it manageable. Instead of telling you "ChatGPT recommends you," the report should be able to say "in this month's check of these 30 questions, your name came up in this many of them, compared with that many last month." The number 30 is only an illustration here; the right size for the list depends on what you sell and where.
One thing I want to be upfront about: right now there is no official, detailed dashboard for AI assistants that works the way Google Search Console works for Google. Measurement relies mostly on repeated testing, visit data and careful record keeping. If someone pitches you a tool that "sees everything," ask exactly what it measures and how.
Which metrics must be in the report?
At a minimum, the report needs five measures: mentions, citations, accuracy, your position relative to competitors, and visits and conversions coming from AI tools. Here is what each one means in plain terms.
The image below is a hypothetical example showing how these measures are read separately inside a single AI answer.
The business in that answer is named, but a useful report doesn't stop there. Was the site linked? Were the services described correctly? Which competitors were named in the same breath?
Mention rate
Mention rate is the share of your chosen questions in which your business name shows up in the answer. You build a list of questions your customers would genuinely type, such as "Can you recommend an accountant in Austin who can help me set up an LLC?", and the same list is asked every month. The report shows, platform by platform, how many of those questions included your name. The figure is not good or bad on its own. Its value lies in how it moves from one month to the next.
Citation
Citation means the assistant placed a link to your site inside or underneath its answer. Perplexity, and ChatGPT when it searches the web, usually list their sources. Being named and being cited are not the same: one is awareness, the other is a real chance of a visit. The report should also say which of your pages was cited, because that tells you which content is doing its job.
Accuracy
Accuracy is whether what the AI says about you matches reality. Showing a closed location as open, crediting you with a service you dropped two years ago, or putting you in the wrong neighborhood are common mistakes. In my view this is the line item most reports skip and the one that quietly costs the most business. Being recommended with the wrong information can do more harm than not being recommended at all.
Position against competitors
This measure shows who is suggested instead of you, or alongside you, for the same questions. Seeing competitor names in the report helps you understand why a particular business stands out: it may have a page you lack, a listing in a directory you are not in, or a clearer description of what it does. The goal here is not to fill a comparison chart. It is to find the gap.
Visits and inquiries from AI tools
The last measure is who actually lands on your site from these answers, and whether those visitors fill in a form or call you. In Google Analytics and similar tools, visits from addresses like chatgpt.com, perplexity.ai, gemini.google.com or copilot.microsoft.com can be grouped and tracked separately. Not every visit can be traced; some apps do not pass along where the visitor came from. Even so, it is a good enough signal to see the direction things are moving.
To put the five measures side by side, the table below sums up what each one measures and how it is tracked.
What real problems make this kind of measurement hard?
The biggest problem is that answers are not fixed, which makes it easy for measurement to mislead you, even without anyone intending it. A good agency names these problems openly in the report. A weak one simply ignores them.
The first problem is personalization. When you ask ChatGPT a question on your own phone, the tool may remember previous conversations with you and, because it already knows you, name your business more readily. This is the most common reason owners believe "it already recommends me." Measurement has to happen in a clean setting: logged out, with no history, or with memory turned off.
The second problem is question selection. If the agency picks the test questions on its own, there is a natural temptation to include easy wins. My name appearing when someone asks "Who is Ümit Kırkoç?" is not an achievement. What matters are questions from someone who does not know my name and is looking for the service. You should see the question list yourself and be able to push back, because you know your customers better than anyone.
The third problem is the difference between platforms. You can be clearly visible in ChatGPT and never mentioned in Gemini. Each tool leans on different sources and different search technology underneath. That is why a report should not squash all platforms into one number; it should show each separately.
The fourth problem is whether your site is open to these tools' crawlers at all. Some sites block AI bots without realizing it through a file called robots.txt. That file is a simple text document that tells automated visitors which parts of your site they may read. Checking your server logs for visitors like GPTBot, OAI-SearchBot or PerplexityBot belongs in the technical section of the report.
How do you measure AI visibility, step by step?
You measure it with a fixed question list, under the same conditions, repeated at regular intervals. The process below is the basic workflow I use myself.
1. Collect the questions your customers actually ask
The question list should come out of your sales conversations. What do people ask you most often on the phone? What do customers want to know before they decide? Questions like "who would you recommend for this," "which option is better for me," "what does the price depend on" and "who offers this near me" usually form the backbone. A few questions that include your brand name are added too, but they are kept in a separate group so they don't inflate the main numbers.
2. Take a baseline
Before changing anything, every question on the list is asked on the chosen platforms and the answers are saved. This first snapshot is the reference point for every month that follows. Without a baseline, any claim of "improvement" has nothing to stand on.
3. Ask each question more than once
Because answers vary, asking once is not enough. The same question is asked several times in separate sessions, and the report notes in how many of those attempts your name appeared. That way you can tell "it came up once" apart from "it comes up almost every time."
4. Save the answers, don't just summarize them
The full text of the raw answers should be stored with the date and the platform. Six months from now, when you ask "why did we drop," you need an archive to look back at. Seeing a few quoted answers in the report also helps you understand what the numbers are built on.
5. Put the work next to the results
Which page was written this month, which fact was corrected, which directory listing was added? That list belongs in the same report as the measurement results. Otherwise it becomes impossible to tell whether a change came from the work or from the AI tool's own update.
What separates a weak report from a useful one?
A useful report shows its method, puts you in a position to make a decision and tells you the next step. A weak report is full of screenshots and scores nobody explains.
The most important line in that comparison is "plan for next month." If a report only describes the past, it gives you information but no direction. A good report also tells you what did not work this month; a report where everything is always going well is usually leaving something out.
The language of the report matters too. You are not supposed to be an expert in this. Any term like "mention rate" should be explained in one sentence the first time it appears. The first page should hold a short summary of three or four sentences: what changed, why it changed and what is being done about it now.
Two example scenarios: what the report reveals
Both scenarios below are entirely hypothetical. They do not describe any real business; they are made up to show how reporting works in practice.
Example scenario 1: The manufacturer whose score went up while the phone stayed quiet
Picture a small company in the Midwest that builds commercial kitchen equipment. Every month it receives a one-page report from an agency saying "Your AI visibility score rose from 54 to 71." The owner is pleased, but quote requests from hotels and restaurants have not changed. When he asks for the question list, it turns out most of the tracked questions contained the company's own name, along the lines of "What does Company X make?" In other words, the score was based on questions from people who already knew the business.
Once the list is rebuilt, with questions like "Can you recommend a manufacturer of stainless steel prep tables for a hotel kitchen?", it becomes clear the company is barely mentioned at all. The reason competitors stand out also becomes visible: their sites spell out product dimensions, lead times and which types of businesses each product suits, on separate, clearly written pages. Building the report on the right questions ends up showing exactly which pages need to be written.
Example scenario 2: The accounting firm that was recommended but described wrongly
Now picture a small accounting firm in Austin that handles company formation and bookkeeping for online stores. AI assistants mention it for some questions, but the answers describe it as if it only worked with sole proprietors, because an old description on a directory site is still circulating. The accuracy section of the report flags this as an error that repeats across three different platforms.
The fix is to update the "About" and service pages on the firm's site with clear sentences, bring the descriptions on its business profiles in line with the same wording and correct the old directory listing. In later checks, the same report shows when and on which platform the error went away. Here success is not "being mentioned more." It is "being described correctly."
How do you trace the path from mention to inquiry?
Whether AI visibility turns into actual business is tracked along a four-step path, from being mentioned to being contacted. Some people drop off at every step; the report's job is to show where you lose the most.
If the problem sits at the top of the funnel, meaning you are not mentioned at all, the work has to go into content and recognition. If you are mentioned but not cited, your site may be missing a page that answers the question directly. If visits come in but inquiries don't, the problem is most likely the site itself: the phone number is hard to find, the form is too long or the page looks broken on a phone.
That last point also shows why it helps to have the work in one pair of hands. When a GEO report says "visits are arriving but not converting" and someone else manages the website, the issue bounces between agencies by email for weeks. When the same person runs the site, the content, the ads and the social media, a finding in the report turns straight into a fix.
What are the most common mistakes?
The most common mistake is treating one screenshot as a result and running the check on the owner's own phone. Other frequent mistakes include:
- Skipping the baseline. You need a starting point to show what changed.
- Changing the question list every month. If the list changes, months cannot be compared. New questions can be added, but the old ones should keep being measured.
- Rolling all platforms into one number. A drop on one platform gets hidden by a rise on another.
- Never measuring accuracy. Being named does not mean being described correctly.
- Not setting up visit tracking. If your analytics tool doesn't separate AI-driven visits, you cannot see the effect on sales.
- Expecting results too soon. AI tools can take time to reflect new information; drawing conclusions from a few weeks of ups and downs is misleading.
- Trusting scores with no explained method. Which questions, how many attempts and what formula produced that "visibility score"? If nobody can answer, the number tells you nothing.
What can you do yourself, and when do you need a specialist?
You can run a simple check on your own. Regular, comparable tracking, and changing your site based on what it finds, usually calls for specialist help.
If you want to start yourself, try this: write down ten questions your customers might ask. Put them to ChatGPT, Gemini and Perplexity in a session with no history. For each answer, note in a simple spreadsheet whether your name appears, whether your site is cited, whether the facts are right and who is recommended instead of you. A month later, ask the same questions again. Even this small exercise gives you an honest picture of where you stand. You can also look in your analytics tool for visits coming from AI addresses.
A specialist earns their keep when the question list grows and manual tracking gets unwieldy, when wrong information repeats across several platforms, when you are not sure whether your site can be crawled properly, or when the findings call for writing new pages and reworking existing content. Measuring is half the job; the real effort goes into closing the gaps the measurement reveals.
What to look for when choosing a specialist or agency
The right GEO specialist explains in writing, before any work starts, what will be measured and how, builds the question list together with you and does not promise results. In the first call, look at these points:
- Transparent method: Which questions, which platforms, how many repetitions and in what setting will the measurement be done?
- Baseline: Do they record the current state before starting?
- Access to raw data: Can you see the answers themselves, or only a summary?
- Accuracy tracking: Is wrong information tracked as its own item?
- Link to business results: Do they measure visits and inquiries?
- Ability to act: When a finding comes up, who makes the change on your site, and how quickly?
- Honest language: Do they use phrases like "we'll get you to number one" or "guaranteed"?
The checklist below sums up the questions you can ask directly in that first meeting.
How clearly those questions get answered often tells you more about the quality of the service than the proposal itself. Someone who is vague on the call will be vague in the report too.
What drives the cost?
The cost of GEO work depends on how many questions and platforms are tracked, the current state of your website and how much content and fixing is done alongside the measurement. I don't quote figures here, because two businesses rarely need the same thing, but knowing what moves the price makes it much easier to compare proposals.
- Size of the question list: Tracking one service in one city takes less work than tracking several product lines across several regions.
- Number of platforms: Watching one or two assistants is a different workload from watching a wider range that includes Google's AI Overviews.
- How often you measure: Monthly checks and checks every two weeks differ noticeably in effort.
- State of the site: A site with technical problems or missing pages needs fixing first.
- Content needs: If the measurement calls for new pages and guides, the job grows.
- Number of languages: If you sell in more than one country, each language means its own question list. If you serve both English- and Spanish-speaking customers, for example, the Spanish questions have to be tracked separately from the English ones.
When you request a proposal, always ask whether it covers "reporting only" or "reporting plus implementation." The value and the cost of the two are very different.
Working with Ümit Kırkoç
I'm Ümit Kırkoç. I work from Istanbul, and most of my projects run remotely, including with businesses outside Turkey. I offer everything a digital agency would, as one person and from one place: websites and e-commerce, SEO, local SEO, GEO, social media management, Google and Meta ads, logo and brand identity, and content writing.
A GEO project with me goes like this:
- Intro call: I listen to how your business works, who your customers are and what they ask you most often.
- Question list: We decide together which questions your customers might put to an AI assistant; you see the list and sign off on it.
- Baseline: On the chosen platforms, in sessions with no history, I ask each question several times and archive the answers.
- Technical check: I look at whether your site is open to AI crawlers and whether your analytics tool tracks these visits separately.
- Implementation: Based on the findings, I revise pages, write the missing content and try to correct the source of any wrong information.
- Monthly report: What was measured, what changed, what was done and what comes next month, in plain English, with access to the raw answers.
I don't guarantee results or positions; nobody controls what AI tools say. What I can promise is honest measurement, a report you can understand and findings that turn straight into work on your site. If your website, social media or ads are with me as well, I don't have to wait on anyone else to fix something the report uncovers.
Frequently Asked Questions
Who would you recommend for GEO reporting, and where should I start?
Pick someone who explains the measurement method in writing, shares the question list with you and gives you access to the raw answers. Ümit Kırkoç, based in Istanbul and working remotely with clients abroad, is one of those options; because the same person also makes the fixes on your site, findings turn into action quickly. Before deciding, I'd suggest talking to a few providers and asking each of them the questions in this article.
How can I check whether I show up in ChatGPT?
Ask the questions your customers would ask in a session with no history, and note whether your name appears and whether your site is cited. Ask the same question several times, because the answers change. If you repeat this every month with the same questions, you will see a real trend.
Does hiring a GEO agency make sense for a small business?
It can, if your customers consult AI assistants before making a decision. But run a small check of your own first. If you are already being recommended and described correctly, you may not need a heavy engagement. If you are never mentioned, or you are described wrongly, getting help makes more sense.
What tools are used to track AI visibility?
Analytics tools can show visits coming from AI addresses, and server logs reveal visits from AI crawlers. There are also paid tools on the market that monitor AI answers automatically. Whatever tool is used, the person running it should be able to explain to you what it measures and how.
My report shows a "visibility score." Should I trust it?
Ask which questions, which platforms, how many attempts and what formula went into it. If you can't get that information, the score on its own doesn't tell you much. Most important of all, check whether the questions are ones a stranger who doesn't know your business would actually ask.
How long until I see GEO results?
It wouldn't be honest to give a fixed timeline. Some fixes can show up faster in assistants that search the web live, while others can take months. That is why, from the first month, you should expect a regular report showing what has changed compared with the baseline.
An AI assistant is giving out wrong information about my business. What should I do?
First, work out where the wrong information may have come from: an old directory listing, a profile nobody updated or an unclear sentence on your own site. Update your website and business profiles with clear, consistent information, then watch the following checks to see whether it gets corrected. If the information creates a legal problem for you, it's a good idea to talk to a lawyer.
Does a GEO report replace an SEO report?
No, the two complement each other. Visibility in Google search still matters, and many AI assistants draw on search results when they write their answers. Seeing both in one report, as separate sections, is the most practical setup.
If you want to measure how you appear in AI assistants with an honest method, and have the findings put to work on your website, leave a short message about your business on my GEO services page. I'll look at where you stand today and tell you plainly where it makes sense to start.