ChatGPT has passed a billion weekly users. Google AI Mode has passed a billion monthly users. Those are enormous audiences, and neither number tells you how many people will visit your website. That is the part missing from a lot of AI search reporting.
We have updated our AI Search & GEO Statistics reference with the latest figures we could verify, the dates behind them and the limits of each study. Updating it also meant correcting our own mistakes. One growth figure had the wrong comparison period. Two citation studies had been presented as if they told the same story. They did not.
We build ContextBolt SEO to help connect the answer an AI gives with the search data you can actually inspect. The useful question is whether the buyers you want can find you, understand the offer and take the next step. A rising industry statistic cannot answer that for you.
- AI adoption, citations and visits measure different things. Report each separately before drawing a business conclusion.
- Zero-click search is broader than AI Overviews. A missing click does not identify what caused it.
- Referral quality depends on the sector. Retail conversion figures are not a forecast for a software company or publisher.
- Use the original source and its date. Our updated statistics page keeps the study samples beside the figures.
- Start with a few real buying questions. ContextBolt SEO can check live answers and retain the findings on your SEO Board, alongside connected search research.
A billion users is context, not your market
OpenAI reported more than one billion weekly active ChatGPT users on August 31, 2026. Google reported more than one billion monthly active AI Mode users on July 22.
Keep the word before “active” attached to each number. Weekly and monthly usage are different measures. The audiences overlap, too, so adding them does not produce a total audience of two billion different people.
ChatGPT is also used for writing, coding, learning and plenty of tasks that do not involve searching for a business. Its whole audience is not the audience for your product category.
A more useful starting point is the question a buyer asks. Someone researching a replacement for a tool, comparing two services or checking a limitation is closer to a decision than someone asking for a general definition. You can write a page that helps that person, then inspect whether AI answers actually use it.
The platform milestone explains why the channel deserves attention. Your customers’ questions decide where to spend the work.
Zero-click does not identify the cause
Zero-click search means a search ends without a click. That includes more than AI summaries. A person might read an ordinary instant answer, change the question, or leave without finding anything useful.
SparkToro’s 2026 analysis of Similarweb data puts the US rate at 68.01% for January through April. The sample covers desktop and mobile web, excluding the Google mobile app.
That is not the share of traffic AI took from websites. It is not even limited to searches that showed an AI Overview.
The distinction matters when your own clicks fall. A lower ranking, a different mix of queries, fewer searches for your topic or a change in the results page could all affect the number. Start by comparing the same pages and queries over comparable periods. Then inspect the results for the questions that changed.
Pew’s browsing study found fewer traditional-result clicks when an AI summary appeared. That is useful evidence of a relationship. It does not prove that the same percentage of your traffic disappeared for the same reason.
Our Zero-Click Report helps identify pages whose click-through rates deserve a closer look. Treat its benchmark as a shortlist for investigation. A benchmark cannot reconstruct the decision every searcher made.
A citation can leave the sale with someone else
An AI citation is a link to a source used in an answer. A brand recommendation is the answer naming a business as a choice. They can happen together, but one does not require the other.
Imagine a buyer asks which software to use. The engine links to your comparison page and recommends the competitor you described most clearly. Your page earned a citation. The competitor earned the recommendation. That example is hypothetical, but it shows why a single “visibility” total can hide the part you need to fix.
Read the answer itself. Check whether your brand is named, what the answer says about it, and where the links go. A count of citations cannot tell you whether the product was described accurately or whether the visitor has a useful next step.
Rankings need the same care. Ahrefs’ March 2026 study found 37.1% of AI-cited URLs ranked in the organic top ten for the same query. The roughly 38% headline uses a broader definition that includes other result blocks. Its parsing also changed since the earlier study.
That is why our reference now shows the current distribution, rather than a clean-looking before-and-after chart. A changed measuring method belongs in the explanation.
For the mechanism behind source selection, see how AI assistants choose citations. The practical task is to make the relevant section accurate and useful enough to support the answer you want a buyer to receive.
Better AI traffic depends on the business
Adobe’s July 2026 report found AI referrals to US retail sites were up 62% year over year. Those visits converted 60% better than non-AI traffic. In travel, AI conversion was about 1% lower than non-AI traffic that month.
All three figures belong together. They show why “AI traffic converts better” is too broad a claim for a business plan.
The 60% is a relative improvement, not a statement that 60 in every 100 visitors bought something. The 62% describes growth from the previous year, not the share of retail traffic now arriving from AI.
A percentage also needs its starting size. If a channel sent 100 visits and later sent 162, that would be 62% growth. This is an illustration, not Adobe’s visitor count. You would still need to know how those visits compare with the rest of your traffic and whether they produced customers.
For your site, put the actual number of qualified inquiries, trials or sales beside the referral count. Three visitors and one sale might be encouraging. It is still a small sample, not evidence that you should move the whole marketing budget.
Keep four rows in your report
A useful AI search report should let you answer four different questions without combining them into one score.
| Measure | Question it answers | Keep with the number |
|---|---|---|
| Brand mentions | Did the answer name us, and what did it say? | Engine, exact question, date and number of checks |
| Page citations | Did the answer link to our content? | Cited URL and the passage it supported |
| Referral visits | Did someone click through to our site? | Source, landing page and reporting period |
| Customer outcomes | Did those visitors take a valuable action? | The action’s definition, count and attribution limits |
The last row is the one I would insist on. A visibility score can rise while the business gets nothing it can measure from the change. That does not make visibility worthless. It means you have not yet established the commercial value.
You do not need a hundred prompts to begin. Pick a small set of questions from real sales conversations, support requests or your search data. Keep the wording stable when repeating a check. Record the engine and date so a change in the test does not masquerade as a change in your brand.
ContextBolt SEO’s AI visibility checks return individual answers from ChatGPT, Gemini, Claude and Perplexity. Historical mention data covers ChatGPT and Google AI. Your findings save to the SEO Board, so you can return to the evidence instead of relying on a screenshot you cannot find.
Keep customer outcomes in your analytics too. Connected Search Console research gives you Google’s search performance; it is not a complete sales-attribution system. The AI visibility guide walks through the checks, while the share-of-voice guide explains why a percentage depends on the questions behind it.
Use the statistics without changing their meaning
If you are writing an article, preparing a client report or asking an agent to research this topic, start with the AI search statistics reference. Each source entry identifies its period and sample. Follow the link to the original research before quoting it.
Keep the country, period and comparison in the sentence. A US retail result from July is not a worldwide ecommerce result for the year. An experimental visibility improvement from 2024 is not a promised gain on today’s engines.
Watch dates especially closely. “Reviewed September 2026” means we checked the material then. It does not turn a 2025 study into 2026 data. Older research can still answer a useful question, as long as the date stays visible.
Then choose one page to improve. Make the answer clearer, correct the product details, or give the reader a useful next step. Repeat the same checks and see what changed.
If you want to do that work with your agent, start a ContextBolt SEO trial. The trial lasts seven days, requires a card and includes 100 credits. After that it is $35 a month for 1,000 credits; cancel before the trial ends to avoid the charge. Start with a real buying question and keep the answer, the citation and the customer outcome separate.
