A lot of listicle content about how search changed in 2026 has circulated this year, most of it quick and shareable. Some of it holds up. Some of it doesn't. I checked each claim against the research actually behind it, corrected the ones with the wrong number attached, and set aside the one I couldn't find any evidence for.

What the data actually shows

1. Search doesn't only mean Google any more, but it hasn't stopped meaning Google either

Semrush's 2026 statistics roundup cites a 2025 industry study finding nearly 35% of US Gen Z use AI chatbots to search for information - alongside YouTube, Instagram and TikTok, not instead of them. The shift toward other platforms is real. A full replacement of Google isn't, at least not yet.

2. Around 60% of Google searches now end without a click

Bain & Company puts the figure at roughly 60%: AI summaries and other on-page features increasingly answer the question directly, before a user reaches a single blue link. That number describes Google's own results pages specifically, not search behaviour generally, so it's worth keeping scoped rather than quoted as a blanket statistic.

3. Unstructured content is the most common reason AI skips otherwise good pages

Research from Princeton and IIT Delhi found that adding specific statistics to an otherwise identical page lifted its visibility in AI answers by up to 34%, and naming a source lifted it by up to 27%. Models extract from the top of a page downward - a strong page with the answer buried under two paragraphs of scene-setting can lose out to a weaker page that leads with it.

4. AI platforms do favour recently updated content

Seer Interactive tracked over 47,000 AI citations across 7,683 pages and found 75% had been updated within the past year - but a quarter of those "fresh" pages were originally published more than two years earlier. The freshness came from maintenance, not from writing something new. More on this here.

5. Visible FAQ content helps. FAQ schema markup doesn't.

This is where one popular claim needs correcting rather than repeating. Ahrefs tested 1,885 pages that added FAQ schema markup and found the code itself made no measurable difference to AI citations. What still works is plain, visible question-and-answer text on the page - every model reads that regardless of what's happening in the markup underneath it. Add an FAQ section because people read it, not because the schema tag is doing any work on its own.

6. Browsers themselves are becoming AI interfaces

ChatGPT Atlas and Perplexity Comet are real, shipped products, not concepts. Comet went free on Mac and Windows in October 2025, Android followed that November, and iOS joined in March 2026 - the full browser is now free across all four with no subscription required. Neither has visibly changed how citation works yet, but a browser with a model built into every tab is a different distribution channel to a results page, and worth watching, not dismissing.

7. AI-referred visitors do convert higher - the real multiplier is 4.4x, not the smaller figures often quoted

Semrush's own research puts the average AI search visitor's value at 4.4 times a traditional organic visitor - based on a study of over 500 digital marketing and SEO-related topics. Two things worth holding onto alongside that number: it comes from one category of topic, not every industry, and AI referral traffic still makes up a small share of total visits across the web. Higher-converting and high-volume aren't the same thing yet.

8. Bing's rankings do shape what ChatGPT cites

A Seer Interactive study, run while ChatGPT Search leaned on Bing's index for real-time retrieval, found 87% of ChatGPT's citations matched Bing's own top results, against 56% for Google. That study is from February 2025, not new this year, but the underlying reason - the OpenAI-Bing retrieval relationship - hasn't been publicly reversed, so Bing Webmaster Tools is still worth setting up, especially since most sites skip it entirely.

9. Off-site mentions beat backlinks, and beat publishing frequency

Ahrefs analysed 75,000 brands across ChatGPT, AI Mode and AI Overviews and found branded mentions elsewhere on the web correlate at 0.66 to 0.71 with AI visibility, against just 0.19 to 0.29 for backlinks. Getting talked about on someone else's site moves the number. Publishing more from your own domain barely does.

10. Citations lean toward business and service pages over blogs and forums

Semrush's breakdown of ChatGPT's sources found business and service websites account for 50% of everything it cites, ahead of news and media sites (9.5%), blogs and content sites (8.3%), and ecommerce (7.6%). A thin company site leaves that ground open for whichever competitor's site isn't.

11. Traditional SEO foundations haven't stopped mattering

None of the above replaces clean technical SEO. Fast pages, sound site structure, clean markup, and content a crawler can actually parse are still the floor everything else sits on - AEO adds requirements on top of that foundation, it doesn't remove it.

12. Updating old content beats publishing new content

Back to Seer Interactive's freshness data: the pages holding citations longest were the ones someone kept current, not the ones published most recently. A well-maintained article from two years ago can outcite something published last week.

One claim I couldn't verify

"Repurpose one piece of content into eight formats to maximise citations" is common advice this year, and I went looking for a study tying it specifically to AI citation behaviour. I didn't find one. What the citation data above actually rewards is depth and originality in a small number of pieces, not the same idea reformatted eight different ways. If a genuine study exists showing format-spreading moves AI citations specifically, I'd be glad to see it and correct this.

A few common follow-up questions

Does this mean SEO and AEO have become the same thing?

No. The overlap is in the foundations - crawlability, page speed, clean structure. SEO ranks a page against a search query. AEO decides whether a model trusts a page enough to quote it in an answer it writes itself.

Of these 12, which matters most if we can only fix one thing?

Structure and freshness together, not one alone. A well-structured page nobody revisits fades. A frequently updated page that's badly structured never gets picked up in the first place.

How do we know if any of this is actually moving our own visibility?

Track it directly, across more than one model. A page can be cited on Perplexity and invisible on ChatGPT, since the platforms don't select sources the same way.