The chart looks like a penalty. Sessions down quarter over quarter. Blog and guide traffic down more than that. So you open Search Console looking for the ranking drop that explains it, and there is no ranking drop. Impressions are flat or up. Positions held. The clicks left anyway.
Stable rankings next to falling clicks is the signature of this change. It is not a penalty and it is not a competitor outranking you. A growing share of your searches are now getting answered before anyone reaches a link.
The click math changed, and it is measurable
Pew Research Center followed the actual browsing of about 900 US adults across 68,879 Google searches in March 2025. When the results page carried an AI summary, 8% of visits produced a click on a traditional search result. Without a summary, 15% did.
Roughly half the clicks, gone. And the summary does not hand them back: links inside the AI summary itself were clicked on about 1% of visits.
The sessions ended differently too. A page with an AI summary ended the browsing session 26% of the time, against 16% without one. The question got answered. There was nothing left to do.
The loss is not spread evenly, and that matters more than the headline
Seer Interactive has been tracking the same thing at brand scale: 53 brands, 5.47 million queries, and 2.43 billion organic impressions from January 2025 through February 2026. On queries where an AI Overview appeared, organic click-through fell from 3.19% to 2.36%, after bottoming out at 1.31% in December 2025.
Now the number almost nobody quotes. On queries with no AI Overview, click-through over the same period went the other way, from 2.8% up to 3.8%.
Two things follow from that pair.
First, this is a reallocation, not a flat decline. The searches losing clicks are the ones an AI can finish on its own. The searches that survived are worth slightly more per impression than they used to be, because the people still clicking are the ones a summary could not satisfy.
Second, your worst hit pages are predictable. Think about the difference between best face serum for oily skin and a search for your brand name plus a size. One is a question. The other is a destination. Questions are what got taken.
For a D2C brand that is a specific and painful inventory: the buying guides, the ingredient explainers, the how to choose posts, the comparison pages. That library was built to catch a question at the top of the funnel, and the question now gets answered upstream. We wrote about where that discovery moved in discovery has left your website.
What teams do next, and why two of the three moves fail
- Publish more. This adds surface area to a channel that is compressing. More guides competing for clicks that a summary already absorbed is effort spent against the trend.
- Move the budget to paid. Not irrational, since paid click-through held up in Seer's data while organic fell. But it swaps an asset you own for rent you pay every month, and your competitors are bidding on the same shrinking page.
- Fix what a machine can read about you. This is the one that compounds, and it is the least popular, because it looks like maintenance rather than strategy.
The foundation did not move
There is a new label for the third option now. AEO, or answer engine optimization. It is worth being precise about what is genuinely new in it, because the parts that are not new are the parts most brands are skipping.
It runs on the same foundation SEO always ran on. A machine has to be able to fetch your page, parse it, work out what it is about, and have a reason to believe you. Every one of those steps predates AI search by twenty years. What changed is the price of skipping one.
Under the old rules, a page a crawler struggled with ranked badly. It still got found eventually, through a link, a brand search, an ad. Under the new rules, a page a model cannot parse is not a lower ranked candidate. It is not a candidate. Adobe found roughly 34% of retail product pages inaccessible to AI systems, and about a quarter of homepage and category content not readable by language models. That is a third of the category handing the shortlist to whoever wrote plainer HTML. We went through the specific failure patterns in a third of retail product pages cannot be read by AI.
What did change: the unit of work
Ranking used to be the deliverable. It is now an input, and a weaker one every quarter.
In July 2025, Ahrefs found that 76% of the pages cited in Google AI Overviews also ranked in the top 10 for that query. Their 2026 study, across 863,000 keywords and 4 million AI Overview URLs, put that at 38%. The rest splits almost evenly: 31.2% from positions 11 to 100, and 31.0% from pages ranking beyond 100.
Their explanation is query fan-out. The engine breaks one question into several smaller ones and pulls sources for each, so the page that answers the sub-question gets cited even though it never won the original search.
The pattern is stronger outside Google. Across 15,000 long-tail queries, Ahrefs found only about 12% of the URLs cited by AI assistants ranked in Google's top 10 for the same prompt.
So a page can be the answer without being the first result, and being the first result no longer books you a seat.
That is harsher in one way and fairer in another. You do not have to outrank a bigger brand to get quoted. You do have to be the clearest available source on one specific question.
Ranking was a position you held. Being cited is a claim you can support.
Three things travel. Everything else stays home.
- Clarity. A recommendation is a restatement, so whatever survives being put in someone else's words is what travels. Premium quality survives nothing. Ships in two days, 90 day returns, free both ways survives every retelling. Write each fact once, in one wording, and use that wording everywhere. If your page says hassle-free returns process and the assistant told the shopper free returns, that shopper has to do a translation step at the exact moment they were closest to buying.
- Trust. An assistant answers with the claims it can point at, and the person receiving that answer goes looking for the same proof. Specifics with something behind them travel further than adjectives: a named standard, a test result, a review count, a third party that said it about you instead of you saying it about yourself. Salesforce found 74% of shoppers say they trust product recommendations from AI chat. That trust gets spent on your behalf before anyone reaches your site, and the page either honors it or burns it.
- Machine readability. The dullest of the three and the one that decides whether the other two ever get read. Facts in the raw HTML rather than assembled after JavaScript runs. Sizing charts and ingredient lists as text, not pictures. Specifications out from behind tabs. Structured data carrying price, availability, and return terms accurately. None of this is new advice. It is a decade old checklist that stopped being optional.
Notice what those three have in common. Not one of them is a trick aimed at a model. They are the same things that help a skeptical human buy faster, which is why the work does not get stranded the next time the platforms change. And they will change: there is no single AI search to optimize for, as the Similarweb data on a fragmenting market showed.
The measurement trap waiting on the other side
Do all of this well and your dashboard can still look flat for a while.
Similarweb followed users who got a brand recommendation from ChatGPT and left without clicking. Those brands were 2.5x more likely to get a site visit in the next seven days than brands the model did not mention. But 55.9% of that traffic arrived through branded search rather than an AI referral link. The conversation that created the intent left no trace, and Google took the credit.
Salesforce measured the same displacement from the other end: between August 2025 and May 2026, discovery on brand-owned properties fell 7% and traditional search fell 15%, while AI assistants and other new channels grew 38%.
So read branded search and direct traffic next to organic sessions, not after them. A branded search line rising while non-branded organic falls is not a decline. It is discovery moving upstream and handing you the buyer one step later than it used to.
The traffic you keep is worth more than the traffic you lost
Adobe Analytics, working across more than a trillion visits to US retail sites, found AI-referred traffic converting 42% better than every other source in March 2026, with 37% higher revenue per visit, 48% longer on site, and 13% more pages viewed.
Fewer visitors, better briefed, checking rather than browsing. That is the trade you are actually in, and it only pays if the page finishes the job the assistant started. More on that reversal in AI traffic went from converting 38% worse to 42% better, and on what gets a brand shortlisted in the first place in what makes an AI agent pick your brand.
Four things to do in the next thirty days
- Split the organic report in two. Queries that are questions, and queries that are destinations. Track them separately from now on. One number hiding both movements is why this shift reads as a mystery decline.
- Ask the assistants about your five best sellers, the way a customer would ask. Every claim that comes back is a claim your page now has to support on sight. Every claim that comes back wrong is usually an extraction problem, not an opinion.
- Read your top pages as a machine does. Fetch the raw HTML and check which facts are missing before any script runs.
- Make six facts plain text on your highest revenue pages: price, availability, sizing or fit, materials or ingredients, shipping time, return terms. Same wording everywhere, in the document and in your structured data.
eLLMo runs test buyers matched to your real customers against your page and returns a ranked list of what stops people from buying, in their own words. A claim a buyer cannot find and a claim a machine cannot extract are usually the same claim, which is why the fix for one keeps turning out to be the fix for the other. For the research on how AI buyers weigh what they read, see AI buyer behavior and structural bias, or see a live run on your own page.
Organic traffic is not disappearing. It is being spent earlier, by a reader that decides in one pass whether you are worth quoting. Clarity, trust, and a page a machine can actually read are the only three things that reader takes with it.
*Related Links: Google users are less likely to click on links when an AI summary appears (Pew Research Center), AIO Impact on Google CTR: 2026 Update (Seer Interactive), Google AI Overview citations from top-ranking pages drop sharply (Search Engine Journal, on Ahrefs data), Only 12% of AI cited URLs rank in Google's top 10 (Ahrefs), Shopping's New First Step: Agentic Search Grows 200% (Salesforce), AI traffic to US retailers rose 393% in Q1 (TechCrunch, on Adobe Analytics data), Discovery has left your website, A third of retail product pages cannot be read by AI.*