You opened Analytics, watched the line slope down for the third month running, and your stomach did the thing. So let’s start with the reassuring part: when your website traffic dropped, the reason is usually something boring. A Google core update reshuffled the deck. A developer shipped a stray noindex tag during a redesign. Your best page slipped from position 2 to position 6. The season turned and your product is seasonal. Any of those will dent your numbers, and every one is easier to find and fix than the scary explanation everyone reaches for first.
That scary explanation is the one you came here about. When Google shows an AI-generated summary at the top of the results, people click through to actual websites noticeably less. Pew Research Center measured it directly: analyzing the browsing activity of 900 U.S. adults, they found users clicked a traditional search result link in just 8% of visits to pages that carried an AI summary, versus 15% of visits to pages without one — nearly twice the click rate when no summary was present (Pew Research Center, 22 July 2025; a snapshot of U.S. browsing in March 2025, and each figure is a share of visits — the study counts sessions, so it describes page-loads more than distinct people).
So yes, AI can absolutely be why your traffic fell. But “AI did it” is a diagnosis you have to earn, not the first thing you reach for. Let me show you how to tell which problem you actually have.
First, read what the data really says
The Pew numbers are the cleanest we have, so it’s worth being precise about them before you panic at a scary headline. Three findings matter for you. First, clicks to outside sites roughly halved when a summary appeared. Second, clicks on the links inside the AI summary were vanishingly rare, happening in about 1% of visits. And third, people were more likely to end their browsing session entirely after a page with a summary — 26% of the time, against 16% without.
The caveats matter too. It’s U.S. adults only, it covers a single month, and it counts visits and sessions, so the percentages describe page-loads more than individual humans. In short, treat it as a strong signal about the direction of travel; it isn’t a national constant you can paste into a forecast.
Here’s the part that should actually calm you down a little: AI summaries don’t show up everywhere. Pew found they appeared on 8% of one- or two-word searches but 53% of searches with ten words or more, and on 60% of queries starting with “who,” “what,” “when” or “why.” Translation: the informational, question-shaped, sentence-length searches are where the summaries live. So if your traffic comes from short, transactional or branded terms — someone typing your name, or “emergency plumber Leeds,” or “buy X size 10” — the AI effect on you is probably small. But if you live on “how to,” “what is” and “why does” content, you’re standing in the blast radius. Knowing which camp you’re in is half the diagnosis.
How to confirm AI is why your website traffic dropped
Don’t act on a hunch. Because the fix for each cause is completely different, spend twenty minutes ruling things out first — guessing wrong wastes months.
Check the shape of the drop. A sudden cliff on a single day usually points to an algorithm update or a technical break; AI erosion, by contrast, is gradual and creeps in over weeks. So line the date up against Google’s confirmed core update history. If your fall matches an update, that’s your lead suspect.
Compare impressions to clicks in Search Console. This is the single most useful test, so do it first. Pull the last 16 months and look at the two lines together. If your impressions held steady or grew while clicks fell, you’re still ranking and still being shown — the click just isn’t happening.
That gap is a strong indicator that something on the results page — an AI summary, a featured snippet or another SERP feature — is resolving the query before anyone reaches you. If impressions dropped too, however, the story is different: you lost rankings, and that’s a separate fight with a separate fix.
Segment by query type. Next, filter your queries into informational versus transactional and look at them apart. If the “how to fix a leaking tap” pages are bleeding while “plumber near me” holds firm, that pattern points squarely at AI Overviews. But if everything is down more or less evenly, look at technical health or a possible penalty instead.
Rule out the self-inflicted wound. Genuinely, check this before anything else. Crawl your site. Read your robots.txt line by line. Confirm your key pages still return a 200 and aren’t quietly carrying a noindex left over from a staging environment. I have lost count of the “Google hates us” emergencies that turned out to be one checkbox flipped during a redesign.
Work through those four and you’ll usually land on a clear answer: algorithm, technical, lost rankings, or genuine AI interception. Only the last one needs the new playbook.
If AI really is the cause, here’s what actually works
The instinct is to write more of the same content, harder. Resist it. If a summary is already answering “what is compound interest” better than your 1,500-word explainer ever will, another explainer won’t win the click back. So the work goes elsewhere.
Move toward questions the summary can’t finish. AI is good at settled, factual answers and weak at judgement, comparison, current pricing, availability and anything local or personal. For example, pages that compare two specific options, show real prices, reflect this year’s situation, or need a human recommendation still earn the click because the summary can’t safely resolve them. Shift your effort down the funnel, toward the searches where someone is close to choosing.
Aim to get cited. If people aren’t going to click through as often, being named inside the answer becomes your new front page. That means leading with the actual answer in your opening sentence, above the 400 words of preamble you’d normally warm up with, and publishing specifics only you have — real numbers from your own work, prices, timelines — because a generic paragraph is already covered by a thousand pages the model has read. And whether ChatGPT, Perplexity or Google’s summary currently mention you is something you can and should measure; we walked through exactly how in our guide to checking your AI search visibility.
Stop depending on a single tap. If one channel’s mechanics can halve your traffic overnight, the real problem is the concentration, not the channel. So build the email list you keep meaning to build. Turn up in the places these systems already trust — Pew found Wikipedia, YouTube and Reddit were the most-cited sources in both AI summaries and standard results, which tells you where the attention pools. A drop like this is a painful nudge to own an audience you don’t rent from an algorithm.
And if all of this feels like it’s asking you to relearn SEO from scratch — it partly is, and it’s worth understanding what genuinely changed versus what’s just new vocabulary. We laid out the difference in SEO vs GEO vs AEO, if you want the map before you spend a budget.
When the honest answer is “hire someone”
Often, frankly. Diagnosing the drop is a fair DIY job — the Search Console checks above fit into one focused afternoon. Acting on the diagnosis at any scale, however, is a different animal, and pretending otherwise doesn’t help you.
If you’re competing in a crowded category, running several languages or markets, or simply don’t have the hours, a good agency or an experienced freelancer will move faster than you will, and it won’t be close. They’ve watched this exact pattern play out across dozens of sites, so they already know which lever tends to shift first. Paying for that instinct is usually cheaper than buying it with your own trial and error.
The one thing worth insisting on: hand them the evidence you gathered. Walk in with “impressions steady, clicks down on informational queries since May” and you’ll get a sharp plan. Walk in with “traffic’s down, do something” and you’ll get a proposal built on guesswork.
Where Hepteon fits
Everything above is a loop: diagnose, respond, measure, adjust. It works. It also tends to die quietly around week three, because you have a business to run and a spreadsheet doesn’t chase you.
That’s the gap Hepteon is built for. It’s seven AI agents that run a website end to end, named after the seven events of the heptathlon — Strategist, Connector, Technical, Writer, Amplifier, Results and Publisher. Between them they watch how your traffic behaves across search and AI engines, catch the technical faults that quietly block crawlers, write toward the searches that still convert, and report what actually moved against the goal you set. As a result, a drop like this one surfaces early, while you can still act on it, long before it becomes the quarterly surprise nobody saw coming. Agencies and freelancers can also run the whole thing white label for their own clients, which applies the same loop across a whole portfolio of sites at once.
But start with the diagnosis. Open Search Console, put impressions next to clicks, and find out whether AI is really your problem before you spend a euro solving it. Half the time the culprit is duller and more fixable than the headlines suggest — and the other half, at least you’ll be acting on evidence.