What the search engine became.

Type a query into Google this morning and what comes back, more often than not, isn't the thing you typed a query into Google for twenty-five years to get. No list of blue links at the top. A paragraph instead, written by a model, sitting where the first organic result used to live.

We didn't quite notice when the search stopped being a search. It happened over a few quarters in late 2024 and through 2025, in the way these category shifts always happen, which is to say without an announcement and with the old vocabulary still in use afterward. People still say "I'll Google it." What they mean now is "I'll ask it." The thing they're asking is no longer the thing they were Googling.

Some numbers, because the numbers are the only thing that settles an argument like this. Roughly 15% to a quarter of Google searches in commercial verticals now lead with an AI Overview (figures vary by audit, and the share is climbing every quarter). Pew Research's 2025 browser study put a number on what buyers do with those Overviews: only 8% of users click a traditional result when an AI summary is present, against 15% when it isn't. Other measurements have the click reduction closer to 58 percent. Methodology accounts for the spread. The direction doesn't move.

The search box did not get better. It became something else. The thing it became is an answer engine, and the answer is rarely written by you.

Meanwhile, on the other side of the same shift, several hundred million people are bypassing Google entirely. ChatGPT, by OpenAI's own disclosure, served roughly 900 million weekly active users by early 2026, more than twice as many as a year earlier. Perplexity processes well over a billion queries a month. Claude with web search, Google's AI Mode, Gemini, Copilot: each one a small Google in its own right, each one returning answers, not links.

Adobe Analytics reports that traffic to U.S. retail sites from generative AI sources rose roughly 1,200% in twelve months. 47% of Gen Z told Adobe last year that they discovered a new brand through ChatGPT in the past twelve months alone.

This is not a modest re-distribution. This is a category change in how buyers find products.

The thing that surprises most merchants, when they finally look, is what AI search actually likes.

Google's AI Overviews are notoriously stingy with retailers. When someone asks an AI Overview for a product recommendation, BrightEdge's holiday 2025 audit found that the model cites a retailer's own pages only about 4% of the time. The remaining 96% comes from somewhere else: review aggregators, magazines, YouTube, Wirecutter, and above all Reddit. The merchant's own product page, the one the merchant has been optimising for fifteen years, is almost invisible to the system that decides what the customer reads.

A separate analysis by Omniscient Digital, of 23,000 AI citations across the major engines, isolated the queries that ask explicitly what people think of a brand. On those, the engines answer with earned media 82% of the time. Brand-owned pages get the remaining 18 between them.

The reason is mechanical, not philosophical. Language models trained on the open web have learned to weight first-person, dated, entity-rich language above marketing prose. There is a useful Princeton/IIT Delhi paper from 2024, "Generative Engine Optimization," that quantified this directly: pages that contain statistics, quoted language, and citations are surfaced in generative answers up to about 40% more often than identical pages without them.

And, in the most counter-intuitive finding of the lot, a 2.4-million-domain study by Writesonic found that the text actually pulled by AI engines into their answers (the cited text) has roughly three times the entity density of normal English. Three times. The sentences these systems lift are stuffed with specifics: brand names, conditions, use-cases, comparisons. The vague consumer-marketing sentence ("the best skin serum on the market") loses to the specific customer sentence ("held up overnight on sensitive skin under SPF") because the second one contains, in machine terms, more answer.

This is the part most merchants haven't sat down with yet.

The single largest reservoir of first-person, dated, entity-dense, use-case-specific language a brand owns is its review corpus. A skincare brand has hundreds of customers explaining, in their own words, exactly the conditions under which the product worked. A homewares brand has photographs, dimensions, and contextual notes ("fits a north-facing kitchen", "softer than I expected"). A supplements brand has the only honest answers to "does it actually work" written by the only people who know.

This is, structurally, the same shape of text that AI engines preferentially cite. First-person. Dated. Entity-dense. Specific to use-case. Difficult to fake. It is the kind of writing the answer engine looks for and almost never gets from the brand's own pages, because the brand's own pages were written by a copywriter and the copywriter was paid to be on-message.

The asset that wins the new search has been sitting on your site for a decade, displayed as decoration.

There is a simple test. Take any product on your store. Open ChatGPT, Perplexity, Google AI Mode, and Claude. Ask each one "is X any good for Y kind of customer?". Note which sources each engine cites in its answer. Almost without exception, the citations will come from somewhere other than your store. They will come from Reddit, from Wirecutter, from Trustpilot, from a YouTube review, from a 2019 forum post.

Your reviews (the verified, dated, signed, first-person ones) are not in the answer. They are stuck in a JavaScript widget that the AI crawlers do not render. They are paginated behind buttons the crawler does not click. They are surfaced as a star count, when the engine is reading for sentences.

The new search rewards the kind of writing your customers have already done for you. The work, in other words, is not to write better marketing copy. The work is to take the writing that already exists, treat it as the asset it is, structure it for the systems that now decide what the buyer sees, and put it on every surface where a buyer can find it.

Reviews stop being something you display. They start being something you publish.

When a customer-success ticket is closed with a careful answer, that answer becomes the page a future buyer reads on Google. When a verified buyer writes the sentence that perfectly captures what your product is for, that sentence becomes the citation in the AI Overview, the line in the next ad, the answer to the question the next ticket would have asked.

This is what the engine underneath a review widget should do, and almost nothing on the public market does.

We have spent the last year building it.

We will tell you more in the summer.