What is AI visibility?

AI visibility is the degree to which an AI assistant names, cites or recommends your business when someone asks it a question. If a customer asks ChatGPT for the best bakery in Austin and three bakeries get named, those three have AI visibility for that question and everyone else has none.

Be aware that the phrase has a second, unrelated meaning. In AI governance and engineering, "AI visibility" means making AI systems transparent and observable to the people running them. A page ranking on Google's first page for this term right now is about that meaning, not this one. This page is about the marketing sense: whether AI answers mention your business.

There is no official definition of either. No standards body publishes one, and neither Google nor OpenAI uses the term in their documentation. Every definition in circulation comes from a company that sells a tool to measure it, which is worth knowing before you treat any single one as authoritative.

Reading these side by side tells you more than any single one. The disagreements are real and they matter when you compare tools.

Three metrics do most of the work, and they are not interchangeable. Mentions count the prompts where your brand is named. Citations count the answers that link your domain as a source. Share of voice expresses your mentions as a percentage of the competitive set for a defined list of prompts. A tool can show a healthy mention count while your citation count is zero, which means AI is recommending you without sending you traffic.

The measurement method matters more than the dashboard. A tool worth paying for sends real prompts to real engines and records the answers. A tool that scores your site's markup and calls the result AI visibility has measured your HTML, not your visibility. Ask any vendor to show you an actual answer about your business before you pay.

The honest caveat almost nobody publishes: these numbers are noisy. The same prompt can return different brands on consecutive runs, so a single check is a sample, not a measurement. Semrush's own research — 230,000 prompts and over 100 million citations tracked from July to October 2025 — recorded ChatGPT's citation of Reddit collapsing from roughly 60% of responses to about 10% within weeks. If the sources an engine trusts can move that far that fast, treat small week-to-week changes in your own score as noise and watch the direction over a month instead.

The reach is real. At Google I/O in May 2026, Sundar Pichai said AI Overviews had passed 2.5 billion monthly active users and AI Mode had surpassed 1 billion. OpenAI said in March 2026 that ChatGPT had more than 900 million weekly active users, and that search usage inside it had nearly tripled in a year.

The click loss is also real, and better evidenced than most claims in this field. Pew Research Center tracked the actual browsing of 900 US adults across 68,879 Google searches in March 2025: when an AI summary appeared, users clicked a traditional result in 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary in just 1% of visits. Ahrefs, studying 300,000 keywords in February 2026, found the presence of an AI Overview correlated with a 58% lower click-through rate for the top-ranking page.

But keep the scale honest. Semrush's traffic study of more than 50,000 websites across 2025 found AI referral traffic grew 66% while still accounting for less than 0.15% of total visits. The near-term risk for most small businesses is not that ChatGPT replaces Google as a traffic source — it is that the clicks you already earn on Google get absorbed by an answer you are not part of.

One reason to care anyway: the traffic that does arrive behaves differently. Similarweb's global ecommerce report covering July 2024 to June 2025 put the conversion rate for ChatGPT-referred ecommerce visits at 11.4%, against 5.3% for organic search. Fewer visitors, further along in deciding.

This is where most advice becomes guesswork, so it is worth separating what is documented from what is merely observed or sold.

Documented by Google, in its AI features and your website guide: its generative AI features are "rooted in our core Search ranking and quality systems." It describes grounding — retrieving relevant pages from the Search index to support an answer — and query fan-out, where the model issues several related searches behind a single question. That last mechanism is the strongest documented argument for covering the cluster of questions around a topic rather than one keyword. Google also states plainly that structured data is not required for generative AI search and that there is no special markup to add.

Documented by OpenAI, in its crawler documentation: OAI-SearchBot is what surfaces sites in ChatGPT's search features, and sites opted out of it will not appear in ChatGPT search answers. GPTBot is for model training, and ChatGPT-User handles user-triggered fetches. These are separate controls — blocking GPTBot does not remove you from ChatGPT search, and blocking OAI-SearchBot does.

Observed but not documented: which sites get cited most. Measurements from Ahrefs in September 2026 and Semrush in late 2025 both put Reddit and Wikipedia at the top of ChatGPT's most-cited domains, with mainstream publishers and review sites behind them. That is a real pattern and a reasonable thing to act on, but it is a measurement of behaviour, not a published rule, and Semrush's data shows it shifting sharply over months.

Speculation, and it should be labelled as such: llms.txt files, content chunking strategies, entity tuning and training-data recency. None are documented by any engine as ranking factors.

Each of these is contradicted by a primary source, not by our opinion.

You do not need to buy anything to get a first reading, and a small business can do this in an afternoon.

Start with Search Console. Google rolled its generative AI performance data out to all websites by 31 August 2026, so impressions from AI Overviews and AI Mode now appear in your Search performance reports. It is impression-side only for now, but it is first-party and free.

Then ask the engines yourself. Write the ten questions a customer would ask before buying from a business like yours, put each one to ChatGPT and Gemini, and record who gets named. Run them twice on different days — the variation between runs is the thing most people miss and exactly why single checks mislead.

Then check you are reachable at all. Confirm your robots.txt allows OAI-SearchBot and the other AI search crawlers, and that no firewall or CDN rule is blocking them, because a site AI cannot fetch cannot be recommended. Only after those three steps is it worth paying a tool to track the same questions continuously.