You already know the frightening numbers. AI Overviews now reach around two billion people a month. Roughly six in ten searches end without a click. Pew Research found that when an AI summary appears, only about 8% of users click a traditional result, down from 15% when there's no summary. Every SEO article of the last eighteen months has quoted these, usually just above a paragraph about how everything has changed forever.
All of it is true, and almost none of it is useful, because the obvious question goes unasked: which traffic did you actually lose? Not how much. Which. Because on that, the data is unusually specific, and it tells a far more interesting story than the panic does.
What AI Overviews actually eat
Look at where AI Overviews appear, according to Semrush's analysis of the queries that trigger them:
→ ~88% are informational. Commercial intent accounts for under 9%; transactional under 2%.
→ ~95% have no ads or very low commercial value. Keywords above roughly $2 CPC remain largely untouched.
→ ~68% get 100 or fewer monthly searches.
→ ~80% sit in the easiest keyword-difficulty band.
Read that list again, and notice what it describes. Low-volume. Low-difficulty. Informational. Commercially worthless. That is not a random slice of the search landscape. It is a precise description of the traffic that never converted anyway.
The "what is a marketing funnel" post. The definitional long-tail article you wrote because a tool said it had a keyword difficulty of 12. The explainer that brought two hundred visitors a month, none of whom ever bought anything, and which you nonetheless put in the quarterly deck because the graph went up.
The reframe AI Overviews didn't kill SEO. They killed the part of your content strategy that was always a vanity exercise. Most businesses lost impressions, not customers, and have been mourning the wrong thing.
And note the shape of the exception: the high-CPC, commercially valuable, transactional queries are the ones Google has largely left alone. That is not an accident, and you don't need to be cynical to work out why. Those queries carry the ads.
The number nobody quotes
Here's the finding that never makes the headlines, because it doesn't fit the story. Semrush's analysis puts the average AI search visitor at roughly 4.4 times the value of a traditional organic search visitor. Adobe's retail data points the same way: visits referred from AI platforms bounce meaningfully less and last considerably longer.
That makes sense once you stop panicking. Someone arriving from an AI assistant has already been pre-screened. They asked a question, got an answer, saw you named, and chose to come anyway. They are further along, better qualified, and closer to a decision than the person who landed on your definitional blog post from a query they'd already half-forgotten.
So the honest summary of mid-2026 is not "traffic is collapsing." It's fewer visitors, worth more each. Which, if you've read our arguments about why zero-click content forces new success metrics, is precisely the trade you should want. Volume was never the point.
One genuine caveat, and it matters: if your business model was impressions, if you're a publisher, an ad-funded site, an affiliate operation, this is not a benign filter. It's an existential one. The optimistic reading here belongs to businesses that sell things, not to businesses that sell attention.
Citation is not a click, and that's the point
Now adjust the mental model. Only around a fifth of users click through to the sources cited in an AI Overview. So if your plan is "get cited, receive traffic," you'll be disappointed, and you'll conclude that AI visibility is worthless. It isn't. You're just measuring it with the wrong instrument.
A citation is brand exposure at the moment of decision. It's closer to being quoted in a newspaper than to ranking in a list. The payoff shows up later and elsewhere: in recognition, in someone typing your name into a search box, in a shortlist you were on without ever knowing it. Which means SEO is quietly acquiring the measurement problems of PR, and the people who insist on last-click proof will spend the next two years defunding the thing that's working.
Two supporting facts worth acting on. Around 70% of people read only the first third of an AI Overview, so being cited late is close to not being cited at all. And over 80% are at least somewhat sceptical of AI answers, with many reporting they've seen inaccurate ones. That scepticism is your opening: ChatGPT users click out to external links roughly twice as often as Google users do, because they are checking the answer. Be the page that rewards the check.
What's actually working
1. Deliberately fish where the Overviews aren't
Given that AI Overviews cluster on informational, low-CPC queries and leave commercial ones largely alone, the strategic move is obvious and almost nobody has made it: reweight your content portfolio toward commercial and transactional intent.
Comparison pages. Alternatives pages. Pricing. "Best X for Y." Buying guides that actually help someone decide. These are the queries where a click still happens because a decision still has to be made, and they're the queries closest to money. The old content strategy of building a huge informational top of funnel to catch people early is the strategy most exposed to AI Overviews, and it was always the hardest to justify anyway.
2. Publish what a model cannot produce
By 2025 a meaningful share of top-ranking pages already showed signs of AI generation, and that share keeps climbing. Which means explanatory content, the summarising, defining, listing kind, is now abundant, cheap and interchangeable. This is the content sameness problem at industry scale, and it has an obvious corollary.
Whatever is expensive to produce is now disproportionately valuable, because scarcity moved:
- Original data. Survey your customers. Publish your benchmarks. A number that exists nowhere else must be cited to be used.
- First-hand testing. Actually use the products. Photograph the results. Models can't try things.
- Real expertise with a point of view. Named humans saying something arguable. Consensus is what AI generates for free.
- Specific, concrete detail. The exact figure, the edge case, the thing that only shows up after you've done it fifty times.
This is why the compounding, expertise-led approach in our content strategy guide has aged better than any keyword-volume playbook.
3. Be machine-readable, or be absent
None of this matters if the systems doing the summarising can't parse you. Clean semantic HTML, content that exists without JavaScript, facts stated as text rather than baked into images, a clear answer near the top of the page rather than buried under six hundred words of throat-clearing. The full picture is in designing websites for AI search, and the broader tactical playbook in our generative engine optimisation guide.
4. Convert much harder
If you get fewer, better visitors, every one of them has to do more work. This is the least glamorous item on the list and probably the highest-return one, because most sites are still built for an era of abundant cheap traffic that could afford to be wasted. Tighten the pages, sharpen the offer, remove the exits, as covered in our landing page design principles.
What isn't working
| The reaction | Why it doesn't work |
|---|---|
| Publish more informational content, faster, with AI | You're mass-producing exactly the content AI Overviews replace |
| Block the AI crawlers | You vanish from the answers and keep losing the clicks |
| Chase a hundred long-tail keywords | That's the band with the heaviest AI Overview coverage |
| Judge SEO on sessions alone | You'll defund the channel producing your best visitors |
| Wait for it to blow over | It won't; AI search is projected to overtake traditional search this decade |
Measure the right things
Rankings and sessions are no longer sufficient, and clinging to them will lead you to correct conclusions about the wrong questions. What to watch instead:
- Which traffic you lost. Segment the decline in Search Console by intent. If informational queries fell and commercial ones held, you're fine. If commercial fell, you have a real problem and it probably isn't AI's fault.
- Whether assistants name you. Ask several of them, in several phrasings, about your category. See if you appear, and whether what they say is accurate.
- Branded search volume. The clearest downstream signal that citations are doing their invisible work.
- Conversions by source. The only number that settles the argument, and the one most teams still can't produce.
If your organic traffic is down and your revenue isn't, nothing is broken. If both are down, the problem is likely further down the funnel than you think, and it deserves an honest diagnosis rather than a panic. Should the answer turn out to be structural, that's what SEO work exists for.
So what do you do
AI Overviews took the informational, low-volume, commercially worthless queries, which is to say they took the traffic that flattered your dashboard and funded nothing. The commercial queries, the ones attached to money, they've largely left alone, for reasons that are not mysterious. Meanwhile the visitors who do still arrive from AI search are worth several times more each than the ones you lost. So the losing move in mid-2026 is to respond by generating more of the exact content that got replaced, faster, with a machine. The winning one is to publish what a model can't: your own data, your own testing, your own opinion, in a form machines can read and humans can verify. Then convert the fewer, better people who show up. SEO isn't dying. It's just stopped rewarding the part that was easy.
Lost traffic, or lost customers?
Find out which queries you actually lost, and rebuild around the ones worth winning.
Explore Search Engine Optimization →Frequently asked questions
Have AI Overviews killed SEO?
No, but they changed which traffic exists. Most queries triggering AI Overviews are informational, low-volume, and carry little advertising value — a precise description of traffic that rarely converted. Commercially valuable queries were affected far less, so most businesses lost impressions, not customers.
Is traffic from AI search worth anything?
Considerably more per visit. Semrush's analysis puts the average AI search visitor at several times the value of a traditional organic one, and Adobe's retail data found AI-referred visits bounce less and last longer. These visitors arrive later in their decision, already pre-screened by the assistant.
Does being cited in an AI Overview bring traffic?
Rarely in volume — only a minority click through to cited sources. Treat a citation as brand visibility rather than a traffic channel. The payoff arrives indirectly, through recognition and later branded searches, which means measurement has to account for mentions, not just sessions.
What content still works in AI search in 2026?
Content a model can't generate: original data, first-hand testing, proprietary numbers, and real expertise with a point of view. Generic explanatory content is now abundant and cheap, while anything requiring actual work has become scarcer and far more likely to be cited.
