---
title: "How to Measure AI Visibility Without Buying a Score"
description: "Build a repeatable panel of buyer prompts, check verified AI crawler fetches, and connect both to enquiries. An in-house method for ChatGPT, Gemini and Google AI Overviews."
url: "https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini"
image: "https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/87188204-5315-4da0-8911-924dc88af017.png"
published: "2026-11-03T06:20:14.247Z"
modified: "2026-11-03T06:20:14.305Z"
---

November 3, 2026 · 9 min read

# How to Measure AI Visibility Without Buying a Score

[Alexander PerelmanHead Of Product @ groas](https://groas.com/author/alexander-perelman)[LinkedIn](https://www.linkedin.com/in/alexander-433793253/)

![A miniature rooftop where a tiny figure records three rain gauges filled to different levels, showing that one reading is weather and the range is what counts.](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/87188204-5315-4da0-8911-924dc88af017.png)

In this article

1. [Why can’t I trust one AI visibility score?](#why-cant-i-trust-one-ai-visibility-score)
2. [Which buyer prompts belong in my panel?](#which-buyer-prompts-belong-in-my-panel)
3. [What can crawler logs tell me that answers cannot?](#what-can-crawler-logs-tell-me-that-answers-cannot)
4. [Did those mentions lead to enquiries?](#did-those-mentions-lead-to-enquiries)
5. [How do I keep this running with one person and a spreadsheet?](#how-do-i-keep-this-running-with-one-person-and-a-spreadsheet)
6. [When does paid tracking earn its fee?](#when-does-paid-tracking-earn-its-fee)
7. [When is a change real enough to act on?](#when-is-a-change-real-enough-to-act-on)
8. [What do I check every Friday?](#what-do-i-check-every-friday)

A single AI visibility score can make your business look absent on Monday and dominant on Friday without anything about your business changing. **You need a repeatable range, not a flattering screenshot.** Run a fixed set of buyer prompts several times, then check two harder signals alongside the answers: which pages AI crawlers fetch and which enquiries mention an AI tool.

That is the in-house system I would build before paying for tracking software. It is deliberately less polished than a dashboard. You can see the prompts, the raw answers and the gaps. More importantly, you can tell the difference between a change worth investigating and another noisy Friday.

## Why can’t I trust one AI visibility score?

Run a prompt once and you have one answer. Run it again and the brand list may change. That variation is normal, which is why [a fixed panel of buyer prompts, run repeatedly and reported with its spread](https://securityboulevard.com/2026/08/the-best-ai-visibility-tools-in-2026-what-each-one-actually-measures/), tells you more than a single percentage. I used to think a screenshot of a good answer made a useful client update. It does not survive the next run.

The word _visibility_ also hides several different measurements. One tool counts answers that mention you; another counts referral sessions; another counts crawler fetches; a fourth estimates an index. [Those methods answer different questions](https://securityboulevard.com/2026/08/the-best-ai-visibility-tools-in-2026-what-each-one-actually-measures/), so their scores are not interchangeable. Even a defined metric such as [Ahrefs’ AI share of voice](https://ahrefs.com/blog/ai-share-of-voice/) depends on its tracked brands, prompt set and weighting. The scope matters as much as the number.

My rule is simple: **if a report will not show its prompts, engines, location and dates, do not use its score to make a decision.**

## Which buyer prompts belong in my panel?

Start with language buyers already use. Pull phrases from sales calls, support tickets and Search Console queries. Look for prices, timelines, problems and comparisons, then write the questions as a buyer would. A Dallas garage door company does not need a prompt about “garage door synergy solutions.” It needs questions such as “garage door spring replacement cost Dallas” and “best garage door company near me with same day service.”

Aim for **24 prompts, frozen for four weeks**. A smaller panel lets one odd answer swing the result; a much larger one makes the routine harder to keep. Changing the questions every Monday is not optimization. It changes the test.

Split the panel into three buckets:

- **Branded: 6 prompts.** Ask about your name, reviews, pricing and service area. This shows what an engine says when the buyer already knows you.
- **Category: 10 prompts.** Use the “best,” “cheapest,” “fastest” and “who should I hire for X in \[city]” questions buyers ask before choosing a provider.
- **Comparison: 8 prompts.** Include “A vs B,” “is \[your service] worth it” and “pros and cons of \[your category].” These put your brand beside alternatives at a decision point.

Run each prompt three times per engine each week across ChatGPT, Gemini and Google AI Overviews. That is up to 216 answers to review. Log whether your business is mentioned, whether your site is cited and where you appear among brands listed. If an Overview does not appear for a run, record that rather than pretending you received an answer.

Report the **mention rate as a range across repeated runs**, by bucket and engine. A 20–45% range is not a tidy 32% visibility score. It is a 25-point spread you need to respect. Keep the raw answers so you can inspect what changed instead of arguing over the average.

## What can crawler logs tell me that answers cannot?

The prompt panel shows what an engine says. Logs show which pages its crawlers request. **A crawler fetch is not a buyer visit, a citation or proof that a particular answer used the page.** [OAI-SearchBot, ChatGPT-User and PerplexityBot requests](https://promvia.app/blog/track-ai-traffic-to-your-website/) are evidence of machine access, not demand. Think of them as delivery trucks at the back door. Useful to count; not a sales report.

Filter first, trust second. [A User-Agent can be faked, so verify crawler IPs against operator-published ranges](https://promptwatch.com/blog/llm-crawler-user-agents-list-2026/). Check Cloudflare or CDN edge logs as well as origin logs: a blocked request may never reach the origin. Then list the pages with verified fetches. Are your pricing, service and comparison pages among them, or are crawlers mostly requesting pages that do little for a buyer?

Do not force a neat story from a log join. In [one analysis of answers and fetches](https://promvia.app/blog/track-ai-traffic-to-your-website/), only 1 of 63 ChatGPT answers had a verified fetch in the preceding 10 minutes, while index crawls preceded 116 of 129 cases. The practical lesson is narrower than “this crawl caused that answer”: **do not expect a fresh fetch immediately before every mention.** Check the top five fetched pages weekly, and use that evidence to decide which pages deserve attention first.

## Did those mentions lead to enquiries?

A mention without an enquiry may still be worth watching. It is not a customer outcome. Start with referrals, then ask buyers directly, because neither view is complete on its own.

[GA4’s AI Assistant channel](https://promvia.app/blog/track-ai-traffic-to-your-website/) captures some AI referrers, but it misses visits that arrive without one. Keep that channel beside a filtered view for `utm_source=chatgpt.com` and session sources containing chatgpt, perplexity, gemini or copilot. For Google AI Overviews, [Search Console’s Web totals do not give you a clean, separate count of clicks from AI Overviews](https://www.cognizo.ai/blog/track-google-ai-overviews-traffic-and-click-loss-gsc-guide/). Do not label pooled search clicks as AI-won traffic.

Then add a _How did you hear about us?_ field. [A comparison of analytics attribution with buyers’ own answers](https://whichaisent.com/blog/how-did-you-hear-about-us-question/) illustrates the problem: someone can copy an answer, search for your name later and arrive with no AI referrer attached. Offer Google, ChatGPT, Gemini, Perplexity, Other AI and Referral, plus an optional “What did you ask?” box. Aim for a 70%+ answered rate and read the free text weekly.

**Keep analytics and self-reported enquiries side by side.** The first shows traffic you can attribute; the second may catch buyers the first misses. Neither should be silently converted into a grand total called _AI visibility_.

## How do I keep this running with one person and a spreadsheet?

You do not need software for the first month. You need a sheet, a calendar invite and the discipline to leave the prompts alone. Give each engine a tab. Put one prompt on each row, record the three weekly runs, and keep the mentioned, cited and rank fields consistent. Run the panel at the same time each Friday, using fresh sessions and the same location setting where available.

Block roughly 90 minutes, but **do not shrink the sample to make the calendar invite look accurate**. Reviewing up to 216 answers takes attention. If the work runs long, keep the schedule and the raw answers intact; those are the parts that make next week comparable. A panel run whenever somebody remembers is a collection of screenshots, not a measurement routine.

![Spreadsheet with repeated buyer prompts showing different brand mentions across three runs](https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/35a3064e-906c-426a-adb2-024b02a68ee9.png)

Archive the answers in a dated folder. Once a month, export verified bot fetches and put the top five fetched pages beside your mention ranges. Use the enquiry responses as a third column of evidence, not as a number to add to the other two. For a closer look at this approach, see [how to build visibility tracking without trusting one screenshot](https://groas.com/post/how-to-measure-your-brand-s-visibility-i-2).

Skip daily checking and the temptation to expand to 100 prompts. More frantic sampling will not make one day’s movement meaningful. **Automate the reminder, not the judgment.**

## When does paid tracking earn its fee?

Paid software earns its fee when it takes over repeatable work without hiding the work from you. [Entry prices cited for visibility tools range from about $29 for 15 prompts to about $398 for Brand Radar, with a median near $99](https://securityboulevard.com/2026/08/the-best-ai-visibility-tools-in-2026-what-each-one-actually-measures/). Scheduled runs, location control and answer archives can be worth paying for once your Friday routine starts slipping or your sheet becomes unwieldy across cities.

I would run the manual panel for a month first. That gives you a prompt set and raw answers against which to judge the tool. Before buying, check crawler access, confirm you can use your own 24 prompts, and make sure you can export every raw answer. [Audit your prompts before you buy](https://groas.com/post/before-you-buy-an-ai-visibility-tool-run). If a product locks the questions or shows only its proprietary score, you are paying to lose sight of the test.

**Skip the subscription until you know exactly which manual job it will replace.** A chart cannot decide which page to fix or whether a change survived repeated runs.

## When is a change real enough to act on?

One good week is weather. I look for a sustained shift in the **full mention range**, not a prettier midpoint. If a bucket moves from 20–45% to 25–50%, then 22–47%, it has mostly wobbled. Do not announce a win or rewrite the site around it.

My working threshold is stricter: look for a range that moves at least 10 points beyond its prior band for three straight weeks, then check whether verified fetches to the relevant pages and enquiries naming AI tools move in the same direction. That is a decision rule, not a claim that any one crawler request produced any one enquiry. If one leg is missing, investigate before calling the change real.

Keep the buckets separate here. A stronger branded result does not mean category buyers have started seeing you. Read the answers that changed, inspect the pages involved and compare the enquiry wording. **Act on a pattern you can inspect, not a score you have to trust.**

## What do I check every Friday?

Keep the same prompts, engines and scheduled hour. This is the short version; the [weekly AI visibility check](https://groas.com/post/the-weekly-ai-visibility-check-14-steps-to-measure-your) has the longer run-through. Work down the list in order, and keep the answer archive open while you do it.

1. Run the 24 frozen prompts three times in fresh ChatGPT sessions; log mentions, citations and rank.
2. Repeat in Gemini with the same location setting where available.
3. Repeat in Google AI Overviews; record runs where no Overview appears.
4. **Calculate a mention-rate range for each bucket and engine.** A lone percentage hides the spread.
5. Export verified crawler fetches from edge logs and list the top five pages.
6. Check GA4’s AI Assistant channel, the filtered source view and Search Console’s Web totals without treating pooled clicks as AI Overview clicks.
7. **Read the how-did-you-hear-about-us answers.** Tag named AI tools and review the optional question text.
8. Archive raw answers in a dated folder; keep the prompt list frozen for four weeks.
9. **Compare three weeks before claiming a change.** Skip this and one good Friday can cost you a month chasing noise.

Keeping the panel is not the hardest part. Fixing what it reveals is. A vendor can show you the prompts where you are absent and leave the page changes, bot-access problems and repeat checks on your desk. That is the gap groas aims to close: continuous execution across search, with a human strategist setting direction and accountability, rather than another score for a client to interpret.

First thing today: open your last 10 closed deals and copy the question each buyer asked before finding you. Turn those into your first 10 prompts, add eight comparisons and six questions using your brand name, then run the list in ChatGPT, Gemini and AI Overviews. Record whether you appear, and mark any run with no Overview. Date the sheet. On Friday, run those **same questions again**. That is where measurement starts.

## Frequently Asked Questions

### Why is a single AI visibility score considered unreliable?

Variation is normal because AI responses change frequently even with the same prompt. Relying on one screenshot or single run does not provide a consistent measure of business visibility. You must use a fixed panel of prompts run repeatedly to understand the true range of visibility.

### What is a practical, low-cost system for measuring AI visibility?

The recommended system involves building a repeatable panel of buyer prompts, verifying which pages AI crawlers fetch, and connecting this data to enquiries. This in-house method does not require purchasing expensive tracking software to begin tracking visibility across major AI tools.

### How should I structure my AI visibility prompt panel?

A recommended panel consists of 24 prompts split into three buckets: six branded questions, ten category questions, and eight comparison prompts. Branded prompts check how engines mention your company, while comparison prompts test your brand against alternatives at a decision point.

### What is the difference between an AI crawler fetch and a real buyer visit?

A crawler fetch is evidence of machine access requesting a specific web page and is not proof of a buyer visit or that an AI answer used that content. Think of these fetches as a technical delivery of information rather than a direct customer demand.

## Paid Ads & SEO, Run For You

Businesses pay one flat monthly fee per service, agencies resell it under their own brand, and groas does the work either way.

[See Pricing](https://groas.com/pricing)

## Structured data

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://graph.groas.com/entity/groas","name":"groas","url":"https://groas.com/","description":"groas is a done-for-you marketing service, not software: elite people working with proprietary technology systems run search ads, Meta ads, SEO and AI search for businesses, and white label for agencies, with a dedicated senior strategist owning every account.","logo":{"@type":"ImageObject","url":"https://groas.com/icon-512.png","width":512,"height":512},"sameAs":["https://www.linkedin.com/company/groas/"]},{"@type":"WebSite","@id":"https://groas.com/#website","name":"groas","url":"https://groas.com/","inLanguage":"en-US","publisher":{"@id":"https://graph.groas.com/entity/groas"}},{"@type":"WebPage","@id":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini#webpage","url":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini","name":"How to Measure AI Visibility Without Buying a Score","description":"Build a repeatable panel of buyer prompts, check verified AI crawler fetches, and connect both to enquiries. An in-house method for ChatGPT, Gemini and Google AI Overviews.","inLanguage":"en-US","isPartOf":{"@id":"https://groas.com/#website"},"about":{"@id":"https://graph.groas.com/entity/groas"},"primaryImageOfPage":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/87188204-5315-4da0-8911-924dc88af017.png"},"breadcrumb":{"@id":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini#breadcrumb"},"dateModified":"2026-11-03T06:20:14.305Z","mainEntity":{"@id":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini#article"}},{"@type":"BreadcrumbList","@id":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://groas.com/"},{"@type":"ListItem","position":2,"name":"Blog","item":"https://groas.com/blog"},{"@type":"ListItem","position":3,"name":"How to Measure AI Visibility Without Buying a Score","item":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini"}]},{"@type":"BlogPosting","@id":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini#article","url":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini","mainEntityOfPage":{"@id":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini#webpage"},"isPartOf":{"@id":"https://groas.com/blog#blog"},"headline":"How to Measure AI Visibility Without Buying a Score","description":"Build a repeatable panel of buyer prompts, check verified AI crawler fetches, and connect both to enquiries. An in-house method for ChatGPT, Gemini and Google AI Overviews.","datePublished":"2026-11-03T06:20:14.247Z","dateModified":"2026-11-03T06:20:14.305Z","image":{"@type":"ImageObject","url":"https://pub-87da24ecbbfc4c3bad6875f3aa013712.r2.dev/generated-images/87188204-5315-4da0-8911-924dc88af017.png"},"author":{"@id":"https://groas.com/author/alexander-perelman#person"},"publisher":{"@id":"https://graph.groas.com/entity/groas"},"keywords":"prompts, answers, visibility, without, score, panel, check, run","wordCount":1909,"inLanguage":"en-US"},{"@type":"Person","@id":"https://groas.com/author/alexander-perelman#person","name":"Alexander Perelman","jobTitle":"Head Of Product @ groas","description":"My background includes Goldman Sachs and Stanford Computer Science.","image":"https://groas.com/media/blog/4f17cac3a81acc224cc1b2cfab8f94bbc295a3089d13e8421125771ef8d4064f.jpg","worksFor":{"@id":"https://graph.groas.com/entity/groas"},"sameAs":["https://www.linkedin.com/in/alexander-433793253/"]},{"@type":"FAQPage","@id":"https://groas.com/post/how-to-measure-your-business-s-visibility-in-chatgpt-gemini#faq","mainEntity":[{"@type":"Question","name":"Why is a single AI visibility score considered unreliable?","acceptedAnswer":{"@type":"Answer","text":"Variation is normal because AI responses change frequently even with the same prompt. Relying on one screenshot or single run does not provide a consistent measure of business visibility. You must use a fixed panel of prompts run repeatedly to understand the true range of visibility."}},{"@type":"Question","name":"What is a practical, low-cost system for measuring AI visibility?","acceptedAnswer":{"@type":"Answer","text":"The recommended system involves building a repeatable panel of buyer prompts, verifying which pages AI crawlers fetch, and connecting this data to enquiries. This in-house method does not require purchasing expensive tracking software to begin tracking visibility across major AI tools."}},{"@type":"Question","name":"How should I structure my AI visibility prompt panel?","acceptedAnswer":{"@type":"Answer","text":"A recommended panel consists of 24 prompts split into three buckets: six branded questions, ten category questions, and eight comparison prompts. Branded prompts check how engines mention your company, while comparison prompts test your brand against alternatives at a decision point."}},{"@type":"Question","name":"What is the difference between an AI crawler fetch and a real buyer visit?","acceptedAnswer":{"@type":"Answer","text":"A crawler fetch is evidence of machine access requesting a specific web page and is not proof of a buyer visit or that an AI answer used that content. Think of these fetches as a technical delivery of information rather than a direct customer demand."}}]}]}
```