Your website is becoming a reference book for robots

Most brand websites were built on the same assumption: a person arrives, and the page's job is to win them over. Big hero image, a line of copy that sounds good out loud, a button. Facts were tucked further down for the few who scrolled.

That assumption is wearing thin. A growing share of the "reading" done on a brand site is done by software that will never see the hero image. It pulls the facts out, compares them with other sources, and writes a short answer for a person who may never visit. Your website is turning into a reference book, and the most frequent reader of reference books is now a robot.

This isn't a prediction. You can see it in the click data.

The click is getting rarer

In March 2025, Pew Research Center tracked the real Google searches of 900 US adults who agreed to share their browsing. Across 68,879 searches, 18% produced an AI summary at the top of the results. Pew then compared what people did next.

No AI summaryAI summary shown

Show the data

Behavior	No AI summary	AI summary shown
Clicked a traditional search result	15%	8%
Ended their browsing session	16%	26%

Source: Pew Research Center, July 22, 2025

When a summary appeared, clicks on regular results nearly halved, and people were more likely to stop browsing altogether. The links inside the summary were clicked on just 1% of visits. Pew's data is from 2025 and from Google only, so treat it as a snapshot rather than a constant. But it's one of the few studies built on observed behavior instead of a survey, and the direction is hard to argue with.

The summaries weren't citing obscure places, either. Pew found 88% cited three or more sources. Wikipedia, YouTube and Reddit made up 15% of the sources cited in summaries, close to their 17% share of standard results. Government sites were cited more often in summaries (6%) than in standard results (2%).

The questions your customers ask are the ones that get answered for them

Here is the part that should make a CPG founder sit up. AI summaries don't appear evenly. They show up most on exactly the kind of searches a curious shopper types.

Show the data

Query type	Got an AI summary
One or two words	8%
Full sentence (noun and verb)	36%
Ten or more words	53%
Phrased as a question	60%

Source: Pew Research Center, July 22, 2025

"Oat milk" is a two-word search, and it mostly still gets a list of links. "Which oat milk has no added oils and works in a latte?" is a question, and it's the kind of search that gets answered on the spot. That's also the version of the search with real buying intent behind it. The people closest to choosing a product are the ones most likely to get their answer from a summary written from your page, or from someone else's page about you.

But people still come to check

If the story ended there, it would be a story about websites becoming pointless. It doesn't end there.

Adobe's July 2026 survey of more than 5,000 US consumers found that 66% of people who use AI to shop say they visit a brand's website directly to verify a recommendation, and 89% say they click on the links or sources an assistant gives them. In the same report, visitors who arrived from AI tools converted 60% better than other visitors (we went through those figures in our data post).

Put the two data sets side by side and a pattern appears. Fewer people click, but the ones who do have already been told about you and are coming to confirm it. They aren't browsing. They're checking a fact.

The machine reads your page to decide whether to mention you. The human reads it to decide whether the machine was right.

So the site now has two readers, and they want the same thing for different reasons.

Two readers, one page

The machine reader wants facts it can quote without risk. It's comparing your page with your Amazon listing, a retailer's page, a review and a Reddit thread. It favors statements that are specific, consistent across sources and easy to find in the page's text. Amazon has said its shopping assistant, Rufus, was trained on the product catalog, customer reviews, community Q&As and information from around the web. Your website is one input among several, and it only counts if it's legible.

The human verifier arrives with a claim in mind: "ChatGPT said this has 12 grams of protein and no seed oils." They want to find that fact in under ten seconds. If they can't, or if your page says something different, you've lost a customer who was nearly sold.

Notice what both readers have in common: neither one is looking at the hero image, and neither one is helped by a slogan. Both are helped by a page that says, in plain text near the top, what the product is, what's in it, who it's for, what it costs and where to buy it.

What a reference-book page looks like

We'd suggest thinking about your most important pages the way an editor thinks about an encyclopedia entry. That's not a call to make them dull. A good reference entry is a pleasure to read precisely because it respects your time.

Shop-window habit	Reference-book habit	Why it matters now
Headline that sounds good ("Taste the sunshine")	Headline that says what it is ("Mango chili dried fruit, no added sugar")	Assistants match pages to questions using the words on the page
Facts in images (nutrition panel as a photo)	Facts in text and tables	Text is what gets read and quoted reliably (tip 14)
Claims without detail ("clean ingredients")	Specific claims with proof ("no seed oils; certified by X, cert no. Y")	Vague claims are hard to quote and easy to skip
Content loaded by JavaScript after the page opens	Key facts in the initial HTML	Some AI crawlers don't run JavaScript (tip 15)
Different facts on site, Amazon and retailers	One set of facts, kept in sync	Conflicting sources make an assistant hedge or pick someone else (tip 5)
One long brand story page	Several pages that each answer one question	A focused answer is easier to cite (tip 6)

None of this means stripping out the brand. The voice, the photography and the founder story still do the work they always did for the human who scrolls. The change is in the order. Facts go first, in text, and the story goes around them.

The uncomfortable trade

Here's what we think is actually new. For two decades, the deal with Google was roughly fair: you let it read your page, and it sent you visitors. AI summaries change the exchange rate. You're still expected to provide the reading material, but fewer visitors come back.

You can respond to that in two ways. One is to treat the machines as freeloaders and hide from them. A few big publishers and retailers are trying that, and for a company whose business is selling the words themselves, it may be right. For a CPG brand, we think it's the wrong instinct. Your product is a jar or a box, not the paragraph about it. If an assistant describes your product accurately to someone who then buys it at Target, you've won, even if your site never logged the visit.

The other response is to become the source the machine trusts most about your own products. That means accepting that much of your site's value will be invisible in your analytics. You'll see it in branded searches, in retailer sales and in the questions customers ask that show they already know about you. Google says it's still sending "billions of clicks to websites every week through AI features in Search." Some of those will be yours if your page is the clearest one on the subject.

A short test

Pick your best-selling product. Ask three assistants a question a real customer would ask about it: "Is [product] gluten free?" or "What's the difference between [product] and [competitor]?" Then read your own product page and time how long it takes to find the answer in text.

Do the same for a competitor's product while you're at it. If their page answers faster than yours, that tells you something about why they keep turning up in answers. Services such as SearchPanel, from XENA Intelligence, will run customer-style questions through the assistants for you on a schedule, but do it by hand at least once. Reading the answers yourself is what changes how you write.

If the assistant got your product wrong, look for the source of the error (tip 25 walks through that). If your own page made you hunt, that's the first page to rewrite. You're not writing for a skimming visitor anymore. You're writing the entry in the reference book, and both of your readers will check it.