Your next customer may never see your website

I run Google Ads and SEO (search engine optimisation, the work of getting a business found on Google) for small companies in South Africa and the Philippines, which means I spend a lot of my week inside other people's analytics. Over the past year those reports have changed shape. Pages hold their rankings and impressions stay healthy, but the clicks that used to follow have gone soft, and referral sources that did not exist two years ago keep showing up.

The plain version of what is happening: a growing share of the research your customers do now ends inside an AI answer instead of on your site. Google's AI Overviews, the machine-written summaries that sit above the normal search results, launched in May 2024 and reached 2 billion users a month by mid-2025, by Google's own unaudited count. ChatGPT has had web search built in since October 2024. When one of these systems answers a question about the thing you sell, it reads your website so the customer doesn't have to.

The neatest illustration of the shift is a prediction that got it backwards in a useful way. In February 2024, Gartner predicted traditional search volume would drop 25% by 2026, lost to chatbots. The volume never dropped. People search more than ever, and Google credits AI Overviews with driving over 10% more queries on the searches where they appear. What collapsed was the click. SparkToro's analysis of search behaviour from January to April 2026 found 68% of Google searches ended without a click to the open web. People kept searching; they stopped leaving the results page.

The best independent measurement comes from Pew Research, which tracked around 68,000 real searches from 900 US adults in March 2025. An AI summary appeared on 18% of those searches. When it did, people clicked a traditional result 8% of the time, against 15% when there was no summary, and clicked a source link inside the summary itself on just 1% of visits. Google disputed the methodology, and SEO tool vendors now claim summaries cover 30 to 48% of queries, though their tracked keyword lists lean towards exactly the questions that trigger them, so I treat Pew's 18% as the honest floor. Publishers, who live on clicks, felt it first: Chartbeat data across more than 2,500 sites showed Google search traffic to publishers fell by a third in the year to November 2025, and the publisher DMG Media told the UK's competition regulator that click-through rates on searches with an AI Overview fell from 25.23% to 2.79%. That last number was produced for a regulatory fight and describes specific result types rather than whole sites, but the direction agrees with everything else.

So how does an AI answer decide whose website to quote? The answer is less mysterious than the industry chatter suggests. When ChatGPT or an AI Overview handles a buying question, it runs ordinary searches in the background, fetches a shortlist of pages that already rank, pulls out the passages that answer most directly, and assembles a summary with citations. That means two things for a business. Traditional ranking still gets you considered, because the shortlist mostly comes from normal search results. And once you are on the shortlist, the pages that win are the ones easiest to quote. A Princeton-led study of 10,000 queries found that adding quotations, statistics and cited sources lifted a page's visibility in AI answers by up to 40%, that keyword stuffing achieved nothing, and that pages ranked around fifth gained the most. It was run on simulated answer engines before AI Overviews even existed, so hold the exact numbers loosely, but the logic matches what I see in client work.

It has changed how I build pages. Every important page now opens with the direct answer in one clean sentence a machine could lift whole. Claims carry numbers and named sources instead of adjectives. Structured data, the machine-readable labels that tell software this is the price, this is the address and these are the opening hours, gets as much attention as the visible copy, and I keep those facts identical across the site, the Google Business Profile and every directory, because a model that finds three different closing times will trust none of them. One thing I don't bother with is llms.txt, a 2024 proposal for a special site map written for AI systems. Google's John Mueller wrote in June 2025 that "no AI system currently uses llms.txt" and compared it to the keywords meta tag, a relic search engines stopped trusting long ago.

The next stage is further along than most owners realise. OpenAI shipped Operator, an AI agent that browses the web and completes tasks on a person's behalf, in January 2025; Perplexity's Comet browser went free worldwide in October 2025; OpenAI's own Atlas browser followed the same month. ChatGPT's shopping results, live since April 2025, are built from structured product data, and OpenAI says they are "chosen independently and are not ads". Adoption is still thin, and I don't know anyone who lets an agent book their plumber yet. But software visiting your site on a customer's behalf reads your markup, checks your prices and moves on; the photography and the brand feeling are wasted on it.

Two things keep me from rewriting the whole playbook. The first is scale. Similarweb counted 1.13 billion referral visits from AI platforms to the top 1,000 websites in June 2025, up 357% in a year, against 191 billion from Google search in the same month. That is under 1% of Google's volume, growing fast, and the visitors who do arrive are engaged, reading roughly twice the pages and staying about twice as long. The second is local. BrightLocal's consumer survey still puts Google as the top platform for evaluating local businesses, used by 83% of consumers in its 2025 edition and 71% in 2026, with people spreading across an average of six review platforms and Apple Maps nearly doubling from 14% to 27%. For the businesses I work with, a workshop in Johannesburg or a clinic in Makati, the Google Business Profile, the reviews and a page that loads quickly still decide most sales. That is still where the budget goes first.

But the direction is set, and it rewards a kind of writing the web drifted away from. The page that wins a citation in an AI answer states its claim plainly, shows its numbers, names its sources and survives being quoted out of context. I find it genuinely funny that after years of SEO tricks, the thing that finally gets rewarded is saying true things clearly. The customer who never sees your website will still be told what it says. What you control is whether the machine doing the telling finds something worth repeating.