Did AI Overviews Kill Your Traffic? What the 2026 Data Actually Shows

· Vittorio D'angiolino
Did AI Overviews Kill Your Traffic? What the 2026 Data Actually Shows

If your organic traffic is down in 2026 and your rankings haven't moved, you're not imagining it and you probably haven't broken anything. Every few months someone publishes a chart showing search traffic falling off a cliff and the marketing internet declares SEO dead. Then the next chart shows things levelling out and everyone moves on. Both are real. Neither tells you what to do on Monday morning.

Here's the version I'd want if I ran a service business. Google isn't deleting your customers. It's redirecting them, and it's doing it in a way that mostly punishes one specific kind of business.

Do AI Overviews actually reduce clicks? What the studies found

The problem with most writing on this is that nobody dates their figures. I went and opened the underlying studies, and the spread between them is enormous, so it's worth knowing which number you're looking at before you panic or relax.

Pew Research Center produced the cleanest behavioural evidence. They tracked 68,879 real Google searches from 900 US adults. When a person hit a results page with an AI summary, they clicked a traditional result 8% of the time. Without a summary, 15%. Clicks on the links inside those summaries happened in about 1% of visits. So roughly half the clicking disappears.

That's correlation though. Maybe people who use AI Overviews were already less clicky. The gap was causal proof, and as of 2026 there is one. Saharsh Agarwal at the Indian School of Business and Ananya Sen at Carnegie Mellon ran a randomized field experiment: 1,065 Chrome users with an extension that randomly hid or showed AI Overviews. Random assignment means the difference is an effect, not a behaviour pattern.

Showing an AI Overview cut outbound organic clicks by 39.8%, from 0.62 per search to 0.37. Zero-click searches rose from 54% to 73%. And for anyone told this is just lazy users getting what they want: removing AI Overviews changed nothing about reported satisfaction, perceived information quality, or ease of finding things. The clicks vanished and nobody felt better off.

Now the longitudinal picture. Seer Interactive has tracked this longer than anyone, and their 2026 study covers 53 brands, 5.47 million queries and 2.43 billion organic impressions from January 2025 through February 2026. Organic CTR on AI Overview queries collapsed through 2025, bottoming out at 1.3% in December 2025, then climbed to 2.4% by February 2026. Seer's own researchers call that levelling off rather than recovery, and say plainly not to forecast from two months of data. I'd take them at their word.

Ahrefs measured the top-ranking result directly: 300,000 keywords, Search Console data. April 2025 showed a 34.5% CTR reduction at position one. By their December 2025 update it was 58%. Whatever else is happening, it hasn't been getting gentler.

And the conservative end, because you should see the range honestly. Amsive Digital's 700,000-keyword study put the average CTR decline at 15.49%, rising to 37.04% when a featured snippet co-occurred. Much milder than Seer. Different methodology, different query mix, different answer.

Which small businesses are most exposed to AI Overviews

The single most useful thing in this whole body of research is that AI Overviews don't fire evenly. Seer's data puts them on about 36% of informational queries and only around 5% of transactional ones. That gap exists for a reason. AI Overviews answer questions, and answering questions is what informational search is for. Transactional searches run on Shopping boxes and map results, which Google can't replace without cannibalising its own ad revenue.

Semrush confirmed the commercial logic directly: 95% of AI Overview keywords have no ads or very low CPC, and keywords above $2 CPC are largely untouched. Google is protecting the queries closest to a purchase. That's not a side effect. It's the design.

Which query types trigger them hardest matters just as much. Comparison queries like "X vs Y" hit 95.4%. Question-format searches, the what/why/how/is/are family, hit 85.9%. Review queries hit 86.3%. And "near me" searches triggered an AI Overview 76.9% of the time.

That last one surprised me. Local intent was always the safe harbour for small businesses, since it runs through map packs and reviews. If nearly four in five local searches show an AI Overview, the harbour has a current in it too.

Pew found the pattern governing all of this. Only 18% of their searches produced a summary at all, but 8% of one or two-word searches triggered one against 53% at ten words or more. The more a person types, the more likely a machine answers instead of you. That quietly inverts long-tail advice: the specific conversational queries a small business was told to target are the ones most likely to be answered before anyone sees a link.

Why "just get cited in AI Overviews" is a trap

The standard response is "write better content, get cited." That's not wrong, exactly, but it has a shape problem.

Seer's finding here is the one everybody quotes, and almost everybody quotes it wrongly. Brand-cited pages get about 120% more clicks per impression than uncited pages on the same AI Overview results page. But the same cited pages still lag 38% behind pages with no AI Overview at all. Being cited doesn't restore your old position. It gets you most of the way back while everyone uncited around you drops through the floor.

Seer's own author also volunteers a caveat the people quoting that 120% tend to skip. Tracy McDonald writes: "We cannot definitively prove that citation causes higher CTRs... it's equally possible that brands with stronger authority and higher baseline CTRs are simply more likely to be cited by Google's AI." What she will say confidently is that cited queries outperform uncited ones regardless of causal direction. If someone sells you citation optimisation on the promise of that 120%, they're selling you a correlation.

So chasing citations means buying more of the thing that just got devalued. You write longer, more structured, more authoritative pages so a model quotes you. If it works, you've spent real money to recover roughly two thirds of a channel your rules for which were rewritten twice in two years. Meanwhile, uncited pages on the same page did worse still.

There's a second trap underneath. The businesses least exposed handle transactional intent, exactly where AI Overviews mostly don't appear. A plumber doesn't get hurt because someone asked Google what causes low water pressure. They get hurt when someone asks for the best plumber near them and Google answers without showing the list. Different problems, different fixes. Treating them as one SEO problem is how people waste a year.

If your site is mostly informational blog posts that exist to attract traffic you then hope converts, you are the most exposed business model in this entire shift. That describes a lot of small service businesses, because "publish helpful articles" has been the advice for fifteen years.

Why your CTR fell but your clicks didn't

Here's something I haven't seen written clearly anywhere, and it matters if you've been staring at a falling CTR in Search Console feeling sick about it.

Seer published a month-by-month breakdown for brand-cited pages in late 2025. In September, those pages got 15.8 million impressions and 398,798 clicks, a 2.52% CTR. In October, impressions doubled to 33.1 million while clicks barely moved, landing at 400,271. Same clicks, twice the impressions, so CTR halved to 1.21%. Not because anything got worse. Because the denominator grew.

November was different, and I'll be honest that it's worse. Impressions rose again to 39.5 million but clicks actually fell to 301,783, and Seer's data can't explain why. That's the month to take seriously.

The practical takeaway:

Check impressions for the same period before you conclude anything. Seer also flags that it can't tell whether the October impression surge came from Google showing more Overviews where brands were already cited, or from those brands genuinely earning citations. Those are opposite stories and the aggregate data can't separate them.

There's a mechanical reason for the whole mess, which is worth knowing before you trust your own dashboard. Google's own documentation says AI Overviews occupy a single position, and every link inside them is assigned that same position. Your reported average position can therefore improve because you're cited in an Overview sitting at position one, while your actual clicks don't move at all.

Rented vs owned: what you can actually control

Here's the reframe I keep coming back to.

Everything above is rented distribution. Google owns the map, Google draws the roads, and Google rewrote the map twice while you were building on it. You can play that game, and for some businesses you should. But you're a tenant, and the landlord just proved he'll renovate whenever he likes.

Email is the opposite.

Nobody between you and the customer decides whether your message arrives based on a model's opinion of your content quality. And the strange thing is that email's rules have been getting clearer and more survivable, which is not the direction most channels move.

Since February 2024, Google and Yahoo have enforced bulk sender requirements on anyone sending 5,000 or more messages a day. The list is short and concrete...

Keep your spam complaint rate under 0.3%, with 0.1% flagged as the line to actually stay beneath. Enable one-click unsubscribe via the RFC 8058 header. Keep valid forward and reverse DNS records.

Notice what that is. It's a checklist. Do it once, your deliverability improves, and it keeps working while you think about something else. Compare that to a channel where the ranking rules change and you re-earn your position from scratch. One is an engineering problem with a correct answer. The other is a negotiation with a counterparty who won't tell you the terms.

One detail worth flagging, because it looks like a trap and I'd argue it's the opposite. Once your domain crosses that 5,000-a-day threshold to Gmail, Google classifies you as a bulk sender permanently. Dropping your volume later doesn't undo it. But the requirements are written down, effective from a known date, and verifiable. There's a document you can read and implement. Compare that to inferring ranking intent from forum posts, where a good outcome carries no guarantee it survives next month.

That's the whole argument. Not "email is back" and not "AI killed SEO." The cost of depending on rented distribution went up, and the cost of depending on owned distribution stayed flat or fell.

How to protect your small business traffic

If any meaningful part of your pipeline starts with a stranger typing a question into Google and picking you off a list, that part is exposed. You can't make the exposure zero. You can stop it being your only route to revenue.

Two things happen in parallel. First, lean into the queries AI Overviews largely leave alone, which means commercial and transactional intent. Someone ready to buy behaves differently from someone researching, and Google treats those searches differently too. Second, capture the people who found you and...

The second one is where most small businesses fail, and they fail it in the same boring way. No capture mechanism, or one nobody fills in, or a list that gets one newsletter and then silence. The plumbing is missing.

The businesses handling this well aren't doing anything clever. A form that converts, a sequence that sends itself, and a way to separate people who asked about one thing from people who asked about another. No agent running campaigns, no sophisticated stack. Systems set up once that haven't broken.

There's a decent chance your list is already sitting there, half-built and unloved. Someone set up a signup form two years ago, it collected a few hundred addresses, the welcome email was never written properly, and nobody's looked since. That's not a strategy failure. It's maintenance debt, and it's cheap to pay off.

What this doesn't mean (and where the data is still uncertain)

Some honest caveats, because I don't want to sell you a clean story.

Seer is explicit that this is levelling off rather than recovery, and that two months of data can't support a forecast. It could be a new normal, a plateau, or the start of something worse. Nobody knows, and anyone telling you otherwise has a product to sell.

The spread between studies is wide enough that you should be suspicious of any single figure quoted without its methodology. Amsive found 15.49%. Ahrefs found 58% on their updated pass. Seer's figures run steeper on informational queries. DMG Media told the UK's competition regulator that click-through rates fell by as much as 89% on their content when AI Overviews appeared. That's a worst case, not an average. Estimates vary because query mix, time period and industry vary. Choose the study that matches your situation rather than the scariest headline.

It also doesn't mean SEO is worthless. If you're cited you get meaningfully more clicks than the uncited pages around you. For businesses where search genuinely is the funnel, competing for citations still pays. The branded exception is worth knowing too: Amsive found branded keywords that trigger an AI Overview actually gain 18.68% CTR. People searching for you specifically still find you.

What it means is narrower and harder to argue with. You're building on rented ground, the rent is going up, and the most reliable hedge is a list of people who already chose you and can be reached without asking permission from anyone.

If your acquisition depends on a stranger typing a question into Google and choosing you from a list, you're exposed. If it depends on a list you own, you're not. That's the whole thing.

Where this leaves you

If you want the short version: search is still worth doing, and it is no longer worth depending on. The businesses that get hurt worst are the ones with a single channel and no list. The ones that shrug this off have customers they can reach directly.

We build automation for exactly that second situation. Post Pilot keeps your social presence consistent without you touching it, and Email Pilot handles the owned-channel side where nobody can change the rules on you. If you'd rather not think about any of it, our done-for-you marketing services cover the whole setup.

Not ready to buy anything? The free tier gets you seven AI-generated posts with no card required. That's enough to see whether the plumbing works before you commit to it.

key takeaways:

faq:

Sources

More stories