GEO

AI brand visibility: how to measure and grow it

How often ChatGPT, Perplexity, Gemini and Google AI Mode name or cite your brand, how to measure it on a schedule, what moves the number and what it costs.

By Ibrahem HelalAI-assisted, reviewed by Jeremy ZavaPublished 23 September 202616 min read

AI brand visibility is how often, and how favourably, assistants such as ChatGPT, Perplexity, Gemini, Claude and Google AI Mode name or cite your brand when a buyer asks a question you should win. You measure it by running a fixed prompt set on a schedule, logging mentions, citations and sentiment, and comparing the result with your competitors.

This is the hub for everything we have written on measuring and growing that number. The work of earning the citations is generative engine optimization; this guide is about knowing whether it worked. If you would rather have someone run the first measurement for you, that is what our AI visibility audit is, and we say so again at the end because it is our own service.

On this page

What is AI brand visibility, and how is it different from ranking?

AI brand visibility is the share of relevant AI answers in which your brand appears at all, and the quality of that appearance. A brand can be named in the answer text, linked as a source, both, or neither. Each of those is a different outcome, and the difference matters more than most reports admit.

A search ranking is a position on a list that everyone sees the same way. An AI answer is generated fresh each time, from sources the engine picked for that one prompt, and it may not include a list at all. So there is no position to hold. There is only a probability that you show up, and a pattern in how you are described when you do.

That changes what you can promise and what you can track. You cannot own the top spot for a prompt. You can raise the share of runs in which you are mentioned, the share in which you are cited, and the share in which the description is accurate and positive. Those three shares are the whole game. The longer argument about why rank and citations drift apart is in AI visibility versus SEO rankings, and the named-or-linked distinction gets its own treatment in AI mentions versus citations.

Google's own guidance draws the line for its features. To be a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to show with a snippet, and nothing more [1]. But being eligible is not being chosen. The answer is assembled by a query fan-out, in which the engine issues several related searches and picks across them [1]. Your page competes in searches you never see, which is why the old habit of checking one keyword tells you so little.

Which AI engines should you measure, and do they cite the same sources?

Measure the engines your buyers use, and treat each one as its own channel. For most business-to-business brands that means ChatGPT, Perplexity, Gemini, Claude, and Google's AI Overviews and AI Mode. Copilot matters if your buyers live inside Microsoft products all day.

They do not cite the same sources, and they do not cite the same number of them. In Orbit Media's tracker of 13,184 citations across four assistants, Perplexity averaged 19.2 sources per answer, Gemini 8.1, ChatGPT 4.5 and Claude 3.6 [2]. An answer that cites nineteen pages has room for you. An answer that cites three or four does not, so each Claude or ChatGPT citation is won against far more competition per slot.

Their tastes differ too. The same study found ChatGPT leaning on primary sources such as documentation, papers and vendor pricing pages, Gemini on Reddit and listicle round-ups, and Claude on review sites and curated lists [2]. Claude did not cite Reddit once across hundreds of answers [2]. Whatever you have read about Reddit being the answer, check it against the engine you actually care about.

Each engine also reaches your site through a different crawler, and you need all of them allowed. OpenAI documents three: OAI-SearchBot for search, ChatGPT-User for actions a user takes inside ChatGPT, and GPTBot for training [3]. Anthropic publishes Claude-SearchBot, Claude-User and ClaudeBot on the same pattern, and Perplexity publishes PerplexityBot and Perplexity-User. Google's AI features use Googlebot and have no technical requirement beyond ordinary indexing [1]. If your hosting or firewall has a block-AI-bots switch, it is quietly removing you from every engine except Google. Our guide to getting cited by ChatGPT goes deeper on the one engine most buyers start with.

How do you measure AI brand visibility?

Write down the prompts, run them on a schedule, and log the same fields every time. That is the whole method. Everything else is tooling.

Start with the prompts. Pick twenty to forty questions a buyer asks at the point of choosing, not at the point of learning: best X for Y, X versus Z, is X worth it, who does X for a company like mine. Orbit Media's tracker runs twelve to forty prompts per client, weekly, which is a sensible band [2]. The way to choose and phrase them, and the traps in wording, is in AI prompt tracking.

Then run them. Fresh session in each engine, no memory, and no logged-in personalisation where you can avoid it. Record whether you were mentioned, whether you were cited with a link, who else was mentioned, what was said about you, and which pages the engine cited. Do the same for three to five competitors, because share of voice is a ratio and the ratio is what moves a budget. AI share of voice walks through the arithmetic.

Then repeat, because one run is noise.

30% of URLs

The choice is between manual and automated, and it is a choice about scale rather than quality. A spreadsheet and a Friday afternoon works for a year, and it teaches you what the engines actually say, which no dashboard does. Tools sample more prompts more often and charge per prompt per engine. Manual versus automated AI tracking sets out when to switch; our rule is to switch once you track more than forty prompts or more than one market.

Which metrics actually matter?

Six numbers, and most reports only need four of them. Mention rate is the share of runs in which your brand is named. Citation rate is the share in which a page of yours is linked. Share of voice is your mentions divided by all brand mentions for the same prompts. Sentiment is whether the description is accurate and positive. Position is whether you are named first, in the middle or last. Source coverage is which third-party pages the engines cite for your prompts, and whether you appear on them.

Mention and citation rates diverge more than people expect. In the same 13,184-citation dataset, Claude mentioned brands in 70% of answers but cited them in 40%, Gemini 72% against 41%, Perplexity 67% against 44%, and ChatGPT 56% against 47% [2]. A brand can be talked about in most answers and linked in a minority. Chasing citations alone undercounts you; chasing mentions alone overstates the traffic you will get.

Resist a single blended score until you understand the parts. Vendors sell an index that rolls mentions, citations and position into one figure, and an index is fine for a board slide. It is bad for deciding what to do next, because a rise in Gemini mentions and a fall in ChatGPT citations can net to zero. AI visibility score explains how those indexes are built so you can read one without being led by it, and AI visibility metrics lists the full set with the formulas and where each belongs in a report.

Sentiment deserves its own line because it is the one that can hurt. An engine that names you and calls you expensive, or out of business, or a subsidiary of someone else, is worse than silence. AI brand sentiment covers how to score it the same way every week so the trend means something.

Why do AI engines name some brands and ignore others?

Two reasons, and they compound. The engine has to know who you are, and it has to find you on the pages it trusts for that kind of question.

The first is an entity problem. An engine that cannot tell your company from another with a similar name, or that has three addresses and two founding years on file for you, will hedge or skip you. Google's Organization structured data exists partly to disambiguate one organisation from another in its results [4]. That job is not done by one tag; it is done by every place your name, site, logo, founders and profiles are listed saying the same thing. Entity SEO is the discipline of making that true, and it is the least glamorous and highest-return work in this guide.

The second is a proof problem. The engines mostly do not cite you; they cite what other people say about you. Claude leans on review sites and curated listicles, Perplexity on LinkedIn and YouTube, ChatGPT on documentation and pricing pages [2]. If you are absent from the review platforms and the round-ups an engine already trusts, there is nothing for it to quote. Review sites AI engines cite names the platforms that matter, category by category.

There is a third reason that is really a version of the first two: nobody searches for you. An engine that has never seen your brand name in a query, a headline or a profile has no reason to reach for it. Press, talks, a founder who posts, and a product people recommend by name all feed the same signal, and none of it is on your website.

How do you make your brand an entity AI engines recognise?

Publish one consistent description of the company in the places machines read, then point those places at each other. In practice that is five moves, in order, and the order matters because later steps use earlier ones as references.

First, the home page carries Organization structured data with the legal name, logo, address, founders and a sameAs list linking every official profile [4]. Google uses those properties to decide which logo and details to show in a knowledge panel [4]. Our guide to organization schema has the full property list and the mistakes we fix most, starting with a sameAs list that points at profiles that no longer exist.

Second, the profiles it points at exist and agree: a company page on LinkedIn, a Google Business Profile if you have a real address, a Crunchbase entry, and the review platforms for your category. Same name, same URL, same founding year, same city, everywhere. A machine reading two versions of a fact trusts neither.

Third, a Wikidata item, but only once there are independent references to hang it on. Wikidata's notability rule requires an item to describe an entity that can be described using serious and publicly available references [5]. An item created with nothing but your own site behind it gets deleted, and a deleted item is a worse signal than none. Wikidata for businesses covers what counts as a reference and how to write the entry so it stays.

Fourth, the Google knowledge panel. Once Google shows one for your brand, claim it, because a claimed panel lets you suggest corrections and choose the featured image. Getting one to appear in the first place is the outcome of the previous three steps plus some press, which is why how to get a Google knowledge panel reads more like a checklist than a trick.

Fifth, keep it maintained. A rebrand, a new office or a founder leaving breaks consistency in a week. Put a quarterly entity check in the same calendar as the prompt runs, and give it an owner.

Entity work feels like admin, and that is why it gets skipped. But every wrong fact an assistant repeats about a client, we have traced to a profile nobody owned. Own the profiles first. The clever content can wait a month.

Where do AI engines get their proof, and how do you get onto those pages?

From lists and reviews, mostly. When a buyer asks for the best X for Y, the engine does not invent a list; it reads several existing ones and merges them. The pages it merges are the best-of round-ups already ranking, the category pages on review platforms, and, for professional services, LinkedIn.

So the work is placement. Find the round-ups the engines cite for your prompts, which is exactly the source-coverage list from your prompt log, and get onto them. Some are editorial and take a pitch. Some are pay-to-play and say so. Some are built from review-platform ratings and require the reviews first. How to get featured in best-of lists has the outreach sequence and an honest view of which paid placements are worth the money.

Review platforms are the other half. Listicle editors often gate inclusion on a minimum review count and rating on the category platform, so reviews come before pitches. Ask every closed customer for a review on the one platform that matters for your category, not five. The platform-by-category map is in review sites AI engines cite.

Your own site still has a role. Publish the comparison and pricing pages the engines want to quote: a real price, a real comparison against named alternatives, a stated method. ChatGPT in particular leans on vendor pricing and help pages [2]. A page that answers how much it costs in the first paragraph is a citation waiting to happen, and most of your competitors will not publish one.

What do you do when an AI answer about your brand is wrong?

Find the source the engine is repeating, fix that source, then make the engine re-read it. Wrong answers are almost never invented from nothing. They are old or wrong pages, faithfully summarised.

Start by asking the engine where it got the claim. Most will show sources, and the wrong fact is usually on one of them: a stale directory entry, an abandoned profile, a press article from a previous strategy, a competitor's comparison page. Fix what you control, request corrections on what you do not, and add the correct fact to the pages the engine already trusts, so the next fan-out finds a majority of right answers.

Then check the crawler side. If an engine cannot fetch your site, it relies entirely on what others say. Confirm your robots.txt allows the search and user-action crawlers for each engine [3], that your CDN is not blocking them, and that the page holding the correct fact is indexed and snippet-eligible for Google [1].

Some errors are opinion rather than fact. Expensive, better for large companies, limited support: these are sentiment, and you shift sentiment by changing the reviews and comparisons the engines read, not by arguing with the model. Fix wrong AI answers about your brand has the full triage, including when a correction request to the engine itself is worth filing, and AI brand sentiment covers the opinion case.

How long does it take, and what does it cost?

Weeks for the entity layer, a quarter or more for the proof layer, and the measurement runs the whole time. Anyone quoting a faster path is describing the easy part.

Entity work is fast because it is mostly within your control. Schema, profile alignment, a Wikidata item once references exist: that is a few days of work and a few weeks for the engines to re-crawl. Knowledge panels are the exception. They appear when Google is confident, and confidence is not on a schedule.

Third-party proof is slow because other people publish it. Reviews arrive at the rate you close customers. Listicle editors update quarterly. Press lands when it lands. Plan on a quarter before the source-coverage number moves and two before share of voice does, and expect the first movement on the engines that cite the most sources, because they have the most slots to fill.

Cost has three shapes. Doing it yourself costs hours: in our experience a thirty-prompt set across four engines, run and logged by hand, is about half a day a week, plus whatever outreach you take on. Tools charge a monthly fee that scales with prompts and engines, and they only measure; they do not do the placement. An agency costs a retainer and should include both the measurement and the placement work, reported together. Ours is on the pricing page, printed, so you can compare it before we talk, and we are biased on this point in the obvious way.

How do you know it is working?

The cited-source list changes before the mention rate does, so watch it first. When the pages you got onto start appearing in the cited URLs for your prompts, the engines have found the proof. Mentions follow, then citations to your own pages, then share of voice against named competitors. Sentiment usually moves last, because it takes a body of consistent descriptions to shift it.

Report monthly, per engine, with the range as well as the average. A report that says the mention rate is 38% without saying it ranged from 25% to 50% across four weekly runs is hiding the most useful fact, which is how unstable the number still is. Stability itself is a result. A prompt that returns you in every run for a month is one you have started to own.

Tie it to something downstream. Search Console reports clicks from Google's AI features inside the ordinary Performance report, under the Web search type [1], so at least one engine gives you traffic you can see. For the others, watch branded search volume and direct traffic: when an assistant recommends you, people go and type your name. And ask every new lead how they found you. An assistant suggested you is now a common answer, and it is data if you record it.

If you want a structured way to check all of this in one sitting, the AI visibility audit checklist is the list we use on the first day of an engagement, and it works just as well without us.

What this guide covers next

Each of the questions above has a longer answer of its own. Read them in this order if you are starting from zero; jump to the one you need if you are not.

• AI visibility versus SEO rankings explains why a page can rank first and never be cited, and what to track instead.
• AI mentions versus citations separates being named from being linked, and says which one you need for which goal.
• AI prompt tracking shows how to choose, phrase and schedule the prompts you measure.
• AI share of voice turns a prompt log into the one ratio a board will act on.
• Manual versus automated AI tracking says when a spreadsheet is enough and when a tool earns its fee.
• AI visibility metrics lists every metric with its formula and where it belongs in a report.
• AI visibility score explains how blended indexes are built, so you can read a vendor's without being led by it.
• AI brand sentiment covers scoring how engines describe you and shifting a description that hurts.
• Entity SEO is about making your company one unambiguous thing to every machine that reads about it.
• Organization schema walks through the properties that matter, the sameAs list, and the errors we fix most.
• Wikidata for businesses covers when you qualify, what counts as a reference and how to write an item that stays.
• How to get a Google knowledge panel explains what triggers one, how to claim it and what you can change once you have.
• How to get featured in best-of lists is about finding the round-ups engines cite and the pitch that gets you onto them.
• Review sites AI engines cite names which platforms each engine trusts, by category.
• Fix wrong AI answers about your brand traces a wrong fact to its source and corrects it for good.
• AI visibility audit checklist is the one-sitting check across crawler access, entity, proof and measurement.

Questions people also ask

• Is AI brand visibility the same thing as GEO?

  No. GEO, generative engine optimisation, is the work of earning a place in AI answers. AI brand visibility is the measurement of whether that work is paying off: how often and how well you appear. You need the measurement before the work, or you are optimising blind.

• How many prompts do I need to track?

  Twenty to forty decision-stage prompts per market is enough to see a trend and few enough to run by hand. Go beyond that only when you cover several markets or languages, and then use a tool rather than a spreadsheet.

• Can I ask ChatGPT about my brand once and take that as the answer?

  No. Answers change between runs and between days, and the sources an engine cites can churn heavily within 48 hours. Run each prompt weekly for at least a month and report the average and the range, per engine.

• Do I need Wikidata and a knowledge panel to show up in AI answers?

  They help and they are cheap once you qualify, but they are not the first step. Consistent profiles and presence on the review sites and lists an engine already cites move the number more, and they are also the independent references Wikidata requires before it will keep your item.

• Will blocking AI crawlers protect my content without hurting visibility?

  Blocking a training crawler alone, such as GPTBot, keeps your pages out of training data while the search crawlers keep working. Blocking the search or user-action crawlers removes you from that engine's answers entirely. Decide per crawler, never with one toggle.

• How much does an AI visibility audit cost?

  It scales with the number of prompts, engines and markets you need covered. Doing the first measurement yourself costs about half a day a week for a month, which is a fair way to learn whether the number matters to your business before you pay anyone, including us.

Sources

1. AI features and your website — Google Search Central (checked 22 September 2026)
2. I tracked 13,184 AI citations across 4 LLMs: here's what I learned — Orbit Media Studios (checked 22 September 2026)
3. Overview of OpenAI crawlers — OpenAI (checked 22 September 2026)
4. Organization structured data — Google Search Central (checked 22 September 2026)
5. Wikidata: Notability — Wikidata (checked 22 September 2026)

Find out what the assistants say about you before your buyers do

Book a free 30-minute consultation with Ibrahem and Jeremy. We look at your site while we talk, and you leave with a prioritised plan.

Book a free 30-minute consultation

Who wrote this

Ibrahem HelalCo-founder

Leads strategy and delivery.

Works on: SEO, GEO

Reviewed before publishing by Jeremy Zava, Co-founder.

Posts by Ibrahem Helal

More on this

• AI Share of Voice: How to Measure ItAI share of voice is the percentage of a fixed prompt set where you're named or linked instead of a competitor, tracked weekly across four assistants.
• AI Visibility Audit ChecklistTen checks that show whether AI crawlers can reach your site, whether your pages can be quoted, and whether buyers hear your name when they ask an engine.
• Entity SEO: A Practical GuideEntity SEO is making your company a single, consistent, machine-readable thing across the web instead of a set of disconnected mentions. Here's how to do it.
• AI Visibility Metrics That Actually MatterCitation rate, mention rate and share of voice measure different things. Here are the AI visibility metrics worth tracking, and how to read them.