Table of Contents
TL;DR
- On May 7 2026, ChatGPT changed how it surfaces brands. Profound measured the share of ChatGPT answers carrying a URL jump from about 4.5% to about 20-24% on a single day, and Similarweb measured ChatGPT homepage referrals rise 354.7% in the week that followed.
- B2B software brands saw daily ChatGPT referrals lift more than 200% across the same shift. E-commerce and retail were essentially flat. The simplest reading is that ChatGPT now recommends companies, not products.
- Earning a brand mention by name is not a ranking exercise. It is a writing exercise. The engine picks the brand whose own pages already wrote the sentence the engine wanted to say.
- Three pages do most of the work for a small business — the homepage, the about page, and one strong category page. Each carries a clean definition of who the business is for, a short proof line, and a few sentences in the buyer’s actual vocabulary.
- The shift is not a guarantee. AI chats cite imperfectly, and a mention by name does not always come with accurate context. The writing discipline is the part that pays off when the engine gets it right.
Most AEO advice in 2026 still tells a small business to chase keyword rankings. ChatGPT does not rank pages. It picks a brand and says its name, and the brand it names is the one whose pages match what the engine wanted to say.
That distinction is the whole game now. A page that ranks well in Google can be ignored by ChatGPT. A page that no one has ever Googled can be the page ChatGPT cites.
In May 2026, Profound’s tracking of the Branded Link Update found the share of ChatGPT responses containing a URL jumped from about 4.5% to about 20-24% on a single day, roughly a five-times lift. Inside the same shift, B2B software brands saw daily ChatGPT referrals more than double their baseline, while e-commerce and retail saw essentially no lift at all. The simplest reading is that ChatGPT now recommends companies, not products.
If that is the new surface, the question for a small business is no longer "how do I rank?" but "what does the engine read on my pages when it decides which brand to name?"
What actually changed on May 7 2026?
For most of the prior year, ChatGPT answered most questions without naming any source out loud. Citations sat behind the text in a footnote pane. The reader who never opened the pane never saw a brand.
On May 7 2026, brand names started appearing inside the answer text as clickable callouts. The reader who never clicked anywhere still saw the names.
That single change shifted the calculus for small businesses. A mention without a click is still a mention, and the reader carries it forward into the next prompt and the next search. The brand that gets named twice across one conversation becomes the brand the reader types into Google an hour later.
Similarweb’s clickstream panel recorded the after-effects across thousands of monitored sites in the week that followed. The pattern is steady, not noisy.
For more on the upstream voice work that makes a brand recognisable inside any extracted answer, see how to fix generic AI website copy.
Why does ChatGPT recommend companies but not products?
The traffic data answers this question directly. In May 2026, Similarweb measured homepage referrals from ChatGPT rising 354.7% week-on-week, and the homepage share of OpenAI referrals jumped from roughly 26-32% before May 7 to about 60% after. The engine sent the traffic to the front door of the business, not to a product detail page.
That is the shape of a recommendation, not a transaction. The engine names a company because a company is what its readers asked about. They asked "what is a good tool for X" or "who can help me with Y", and the engine returned a short list of names plus a description plus a homepage link.
A product is a different question. A buyer who already knows what they want goes to a comparison page, a category page, or a search box. The engine sees no need to pick one product over another inside the answer text, because the buyer will pick later.
So the businesses that gained were the ones whose value proposition reads as a company-level answer. The businesses that did not gain were the ones whose pages read as a catalogue. The first kind of writing earned the named mention. The second kind earned a click somewhere later, if at all.
Which signals does an AI chat use to pick the brand it names?
An AI chat does not read every page on your site before it answers a question. It reads a small set of candidate pages drawn from its index plus a few live retrievals, and it picks the sentence inside those pages that sounds most like the answer it wanted to give.
That extraction step is the chokepoint. The engine extracts a 40-60 word block from one page and frames it as the answer. The brand on that page is the brand the answer cites.
So the question of "what signals does the AI use" reduces to a different question. Which of your pages contains a sentence that reads as a clean, confident answer to a real buyer question?
A handful of moves make those sentences easier to find. The page opens with a one-line definition of what the business is, named in plain terms. A supporting paragraph carries one quantified proof point. The headings on the page read as questions a buyer might type into the chat itself.
Anything that disguises the answer (a long brand vignette, an industry-jargon hero, a parade of feature blurbs) makes the extraction harder. The engine moves to the next candidate page, and the next candidate brand.
How do you write your own pages so the AI can pick you?
Three pages do most of the work for a small business in 2026. The homepage carries who the business is for. The about page carries who runs the business and why. One strong category or service page carries the proof the buyer needs to recommend the business to a friend.
Each of the three is short. Each carries a clear definition. Each names one specific buyer and one specific outcome, in the buyer’s own vocabulary.
The work upstream of the writing is the part that takes time. You read thirty real customer messages and write down the recurring phrases. You name the position your business occupies (who you serve, who you do not, what they otherwise would have done). You write five sentences on how the brand sounds when it talks, and three on how it does not.
Once those three small documents exist, the pages write themselves. The AI tool you use to draft them inherits the substrate, and the page that ships carries the brand’s specific fingerprints rather than the median register of the training data.
The same edit that makes the page sound like one specific business also makes it extractable. A sentence the engine wants to quote is, by definition, a sentence that named one thing clearly. For more on the editing pass that protects that voice through the drafting workflow, see how to edit AI-generated content.
Which move on this list matters most for a small business?
If you can only do one thing this month, do the homepage hero. The hero is the sentence the engine reads first when it lands on your domain, and it is the sentence a busy buyer skims when ChatGPT sends them through.
The hero carries three jobs in twenty words. It names who the business is for, what the buyer gets, and what the buyer otherwise would have done.
The hero that does those three things is the hero that earns the brand mention by name.
A hero that decorates instead of defines (a slogan, a tagline, a poetic line) loses both audiences. The buyer scrolls. The engine moves on. Neither bothers to extract.
A hero that overpromises (a superlative, an industry-leader claim, an unverified number) loses worse. The engine has learned to discount that pattern. The buyer has learned the same pattern faster.
Writing the hero well is the same work that fixes the rest of the page. For the simple structure the hero must carry, see how to write honest headlines.
The other moves all matter. The hero is the one that pays back the fastest, because a single line moves the whole rest of the page.
Other questions worth answering
How reliable is the context that arrives alongside a quoted citation?
Tow Center reported error rates above 60% for some news-query categories, with one ChatGPT configuration surfacing more fabricated than real references. So a callout that names your firm does not guarantee the surrounding sentence describes you accurately. The honest posture is to write copy that earns the citation when the engine gets it right.
How do Perplexity and Gemini approach extracted answers differently from OpenAI?
ChatGPT leans encyclopedic. Perplexity over-indexes on Reddit-style community sources. Google AI Overviews stays the most selective, citing mainly top-ranked organic results. Microsoft Copilot rewards table-shaped content.
The honest read is that one extracted block cannot satisfy every engine at once.
How long after publishing should I expect an engine to surface a citation?
OpenAI surfaces new content irregularly. ChatGPT does not search on every prompt and cites only a portion of the pages it retrieves. So the realistic timeline ranges from days, if the prompt happens to trigger search and your slug sits in the Bing index, to indefinite. Bing Webmaster indexing matters more than a calendar count here.
What conversion rate does OpenAI’s referral traffic deliver to publishers?
Similarweb’s May 2026 clickstream measurement found OpenAI referral traffic converting at 7.1%, second only to paid search across the publishers tracked. Average views per visit rose from 3.8 to 4.7, a 24% lift. The visitor who arrives via a quoted answer reads further and converts more readily than a typical organic visitor.
How would you earn your first AI mention by name?
Pick one buyer question your business is the right answer to. Write it down in the buyer’s own words, the way the buyer would type it into a chat.
Open your homepage and read it aloud against the question. Does the first paragraph answer it? Does any sentence on the page answer it cleanly enough that ChatGPT could copy that sentence into a reply without losing meaning?
If the answer is no, the rewrite is small. You replace one paragraph, add one definition, and name the position in one sentence.
You ship the page and let the engines re-read it on their next crawl.
If your homepage reads as a category description rather than as a clean answer to one named buyer question, you can contact me here. Send the page and the buyer question you would like AI chats to send you. I will run the read-aloud diagnostic and name the one paragraph that is holding the rest of the page back.
No charge and no follow-up call. The page that ships after the diagnostic is the page that earns the first named mention.