AI Search Statistics for 2026, With a Source on Every Number

How many businesses are invisible to AI search?

Most of them, on every measurement anyone has published so far. The two headline figures below were produced independently, by different teams, using different methods on different populations, and they point the same way. That agreement matters more than either number on its own, because a single study can be an artefact of its own sample while two rarely are.

Do AI citations transfer between engines?

Barely, and this is the best-evidenced claim on the page. Two studies run months apart, on different populations, with different methods, landed within half a percentage point of each other. If you take one thing from this page, take this: a visibility score that blends the engines together hides the specific engine you are losing.

What kind of sources does AI actually cite?

Not the institutions you would expect. On real estate marketing questions the engines lean overwhelmingly on ordinary company websites, and a single video platform outranks every major property portal combined. Read the scope note before quoting this one, because the questions behind it are agent-facing rather than consumer property search, and that changes which sites could appear at all.

What is AI search doing to clicks?

Reducing them, measurably, though the picture is less uniformly bleak than most headlines suggest. One of the four numbers below has been recovering since its December 2025 low. Note also the last figure, which is the reality check: the surface everyone is writing about still accounts for a third of one percent of actual searches.

Can AI crawlers even reach your site?

Not always, and the failure is completely invisible from a browser. This is the statistic most likely to apply to you without your knowing, because it is decided above your own configuration and leaves no trace in your own server logs. It also has the cheapest fix on this page, which is why it is here rather than buried in a technical post.

How to use these AI search statistics without getting burned

Three failure modes account for almost every misquoted number in this field, and all three take a minute to avoid. Check the date, because a figure measured before AI Overviews rolled out describes a different product. Check the population, because an ecommerce sample says nothing reliable about real estate. Then check whether the number has a source at all.

The last one is more common than it should be. While rebuilding this page we dropped roughly half the statistics from its previous version for exactly that reason, and the version you are reading is shorter and more useful because of it.

Where these AI search statistics come from, and what is missing

Ten distinct sources, ranging from September 2024 to September 2026, seven of the figures our own and the rest other people’s work linked back to them. Our own measurements, led by the 150-query real estate AI search study, publish their row-level data so you can recompute rather than trust, and our AI citation methodology states what every one of those numbers must carry. The third-party figures come from named studies with stated sample sizes rather than vendor marketing pages.

What is deliberately missing is anything we could not verify. That includes several widely repeated figures about AI search adoption that trace back to a press release with no method attached, and one about citation share that we could only find quoted second-hand. If you have a primary source for either, send it and it goes in. Journalists who want our first-party findings in citation-ready form will find them, with each study’s method and the chart files, in the CiteVantage press kit.

The number that actually matters is not on this page

Market statistics tell you the weather. They do not tell you whether your own roof leaks. Of the businesses in our 181-brand sample, most were absent from AI answers while being entirely convinced their marketing was working, because nothing in their analytics reported the absence. There is no dashboard for an answer that never named you. The 500-query Houston real estate AI visibility study shows the same blind spot at city scale: across 2,000 answers from four AI surfaces in September 2026, about 94% of the Houston Association of Realtors’ 50,000-plus members were never named.

Running the check yourself takes a few minutes: ask each engine the questions your buyers actually ask, in a logged-out session with memory disabled, and write down who gets named. If you would rather we ran it across ChatGPT, Perplexity, Gemini and Copilot and sent you the results, our Free AI Visibility Audit does that inside 48 hours.