As of September 2026, most brands are close to invisible inside AI answers. Webflow analyzed the websites of 2,000 companies and found the median company appeared in just 16% of the AI answers it would want to be part of. Being cited in those answers happened only 6% of the time.
That gap is the whole story. Your buyers now ask ChatGPT, Perplexity, Claude, and Google AI Overviews the same questions they used to type into a search box. The answer comes back with a short list of sources. If your site is not one of them, you were never in the room.
We build websites for B2B teams, and this is the question clients open with now. Not rankings. Citations. So we want to walk through what the current numbers actually say, and what we think is worth doing about them.
AI answer visibility is how often an AI system names your brand or links to your site when it answers a question in your market. It has two halves. A mention is the model saying your name in the text. A citation is the model linking one of your pages as a source.
The two behave differently. A mention can come from the model's training data, from a Reddit thread, or from a review site like G2. A citation almost always means a live crawler fetched a real page and judged it useful enough to point at. Citations are the harder half, and they are the half that sends you traffic.
This matters for how you measure. If you only track brand mentions, you can look healthy while sending no visitors anywhere. If you only track referral traffic, you miss the buyers who read your name in an answer, remembered it, and typed your domain in directly a week later.
Fewer than most teams assume. In the study Webflow published alongside its Conf 2026 announcements in September 2026, the median company across 2,000 analyzed websites appeared in 16% of the AI answers it would want to be part of, and was cited in those answers only 6% of the time.
Read that as a floor, not a ceiling. It means roughly five out of six relevant questions in your category get answered without you. It also means the bar is low. Most of your competitors are not doing this well either, which is unusual for a channel this important.
Webflow also reported that customers using its AEO product saw 70% more mentions and 120% more citations than non-Webflow brands. We would treat that one carefully. Webflow labels it a directional finding from its own internal data, comparing 158 customers who accepted recommendations against 187 who did not, between May 19 and July 13 of 2026. It is a vendor measuring its own product, and the company says plainly that it does not isolate the product from other causes.
A mention costs the model nothing. A citation is a bet. The model is telling the reader that this specific page backs up the claim it just made, which means the page has to be fetchable, readable, and clearly on topic. Broken rendering or a vague page loses the citation.
This is why we keep pushing clients toward static, prerendered HTML for content pages. If the words only appear after JavaScript runs, a crawler that does not execute scripts sees an empty shell. Our own blog is prerendered at build time for exactly this reason, and we treat that as a baseline rather than an optimization.
It is also why page structure matters more than page length. Models extract passages, not whole documents. A section that answers its own heading in the first forty words is easy to lift into an answer. A section that warms up for three paragraphs before saying anything is not. Our guide on how to get cited by AI search engines goes deeper on that structure.
Enough to show up in your analytics. The Pew Research Center published a study on July 22, 2025 that tracked the browsing of 900 US adults across 68,879 Google searches from March 2025. When an AI summary appeared, users clicked a traditional search result in 8% of visits. Without a summary, they clicked in 15%.
That is close to half the clicks, gone. And the AI summary did not make up the difference. Pew found that users clicked a link inside the AI summary in just 1% of visits to pages that had one.
We think the honest reading is that AI answers are compressing the middle of the funnel rather than replacing it. People still buy software. They just arrive later, with more of their research already done, and with a shorter list of vendors they are willing to consider. Getting onto that list is now the job.
Structure and evidence, in that order. The research paper GEO: Generative Engine Optimization, by Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan and Ameet Deshpande, tested optimization methods against a benchmark the authors built called GEO-bench. They report that these methods can boost visibility by up to 40% in generative engine responses.
The methods that performed best were unglamorous. Citing sources. Adding real statistics with the source named. Writing with a clear, authoritative voice instead of hedging. Improving plain readability. None of that requires a new platform or a new budget line.
Keyword stuffing went the other way and reduced visibility. That is worth sitting with. In classic SEO, stuffing was merely useless. In AI search it appears to be actively harmful, because the model is judging whether a passage reads like a trustworthy source and repetition reads like spam.
No, and we would push back hard on anyone selling that. The pages that get cited are almost always the same pages that were already good: clear definitions, real numbers, honest comparisons, current dates. Most sites do not need new content. They need their existing content restructured so the answers are findable.
Start with the pages you already rank for. If a page sits in Google's top twenty for a real question, it is already in the pool that Google AI Overviews draws from. Reformatting that page so each heading is a question and each answer opens the section is a one hour edit with a real chance of paying off.
The pages we would leave alone are the ones with no search demand behind them. Restructuring a page nobody asks about does not create demand. It just makes a quiet page slightly tidier.
Start manually, because it costs nothing and the results are usually blunt enough to act on. Write down the twenty questions a buyer would actually ask before choosing a vendor in your category. Run each one through ChatGPT, Perplexity, Claude, and Google. Record whether you appear, whether you are linked, and who got cited instead.
Do it once and you have a baseline. Do it monthly and you have a trend line, which is what you actually need to tell whether a change worked. Paid monitoring tools exist, and the SEO platforms most teams already pay for, Semrush and Ahrefs among them, are moving into this space, but a spreadsheet and an hour beats a tool nobody opens.
Pair that with server side signals. OpenAI documents three separate crawler tokens: GPTBot for training data, OAI-SearchBot for surfacing sites in ChatGPT search, and ChatGPT-User for actions a person triggers in a chat. Perplexity, Anthropic and Google publish their own tokens too. If those crawlers are not reaching your key pages, no amount of content work will help. Our walkthrough on how to measure AI search traffic covers the analytics side in more detail.
Less than you might expect, which is the good news. Across the projects we have delivered, the things that make a site fast and readable for people are the same things that make it legible to a crawler. Clean semantic HTML. Real headings. Content in the initial response, not assembled later by script.
Where it does change our thinking is in what we ship on day one. We now treat prerendered content, a sensible robots policy for AI crawlers, and clean article markup as part of the build rather than a later SEO phase. Retrofitting those into a live site is slower and more expensive than building them in.
We have also stopped treating the blog as a separate thing from the marketing site. For a B2B company in 2026, the article that answers a buyer's question well is often doing more sales work than the product page it links to. On one of our client projects, Offbeat Travel, the organic traffic gain after rebuild was 190%, and the pages doing that work were content pages, not landing pages.
Pick your twenty buyer questions and run them through ChatGPT, Perplexity, and Google this week. Write down where you appear and where you do not. Then take the three pages closest to those questions and restructure them so each section answers its heading in the first forty words.
That is a genuinely small amount of work, and it is the highest leverage version of this we know. Everything else, the tooling, the dashboards, the monthly reporting, only matters once you know which questions you are losing. If you want a second opinion on which of your pages are worth the effort, or you want us to look at whether your site is even readable to these crawlers, we are happy to walk through it. You can find us at phoenix.studio.
Tell us where you want to go. We'll tell you how we'd get you there.