Traffic fell, the chart looks ugly, and someone just said "it's probably the AI summaries." They might be right. The summaries are a popular suspect because they're genuinely guilty a lot of the time. A wrong diagnosis costs you in both directions, though. Blame the Overviews for an algorithm hit and you'll skip the audit you need. Blame your content for an absorption problem and you'll rewrite pages that were never broken.
So here's the ordered checklist for the afternoon after the drop. Each check comes with its method, so your verdict on the AI search summaries traffic drop rests on evidence you collected instead of a hunch someone said out loud.
Check 1: confirm the drop is real
Start with the boring stuff, every time. First make sure your analytics tags still fire. A consent-banner update or a tag-manager change that silently halves "traffic" is a classic false alarm. Then compare like-for-like ranges, which means this month against the same month last year wherever seasonality exists.
Next, look at whether the decline is sitewide or concentrated. An overnight cliff across the whole site usually means something mechanical broke, while a slow bleed on one page type points to something in the environment. If your clicks fell to almost nothing, work through our checklist for near-zero Google clicks before anything else.
For a quick cross-check, put Search Console's totals next to your analytics totals. If they disagree sharply, your problem is measurement, the cheapest outcome you could hope for.
Check 2: isolate where it lives
Before you name a suspect, split the decline by page and by query. Then label every losing query as informational or commercial. That one split carries most of the diagnosis. Answer absorption lands almost entirely on informational queries, the definitions and how-tos, explainers and glossary lookups. Commercial and navigational queries mostly keep their clicks.
So a drop spread evenly across both intent bands points away from AI summaries and toward an update or a site problem. One practitioner thread put the whole method in a single line: isolate before you speculate.
You're getting AI traffic. But do you know where it comes from?
RankControl credits every visit to the assistant that sent it: ChatGPT, Perplexity, Claude, Gemini, Copilot, or Grok. Full source attribution, next to your Google traffic.
Check 3: read CTR at stable position
Now for the signature test. Pull average position and click-through rate for your losing queries, across the windows before and after the drop. The absorption fingerprint is precise. Position stays stable, impressions hold or rise, and clicks and CTR fall. You're still ranked and still shown, and the searcher got satisfied above you.
Here's a live case with exactly that shape, a 50% overnight click drop while impressions held, on informational queries that already sat under Overviews:
Drop in clicks while impressions grow. AI overview impact or something else?
https://imgur.com/v6ppGnH About 3 weeks ago I noticed a drop in traffic. Clicks declined about 50% overnight, but impressions were holding steady. The trend has continued for the past 3 weeks. These are informational queries and a lot of my...
If position fell along with clicks, stop blaming the summaries. Something moved your rankings, and that calls for a different checklist entirely. And if CTR fell at stable position on commercial queries too, look for SERP-layout changes, new ad slots or a competitor's rich results before you reach for AI absorption.
Check 4: verify the Overview is actually there
This is the step people skip, usually because they once saw an Overview on their own phone. One sighting is an anecdote. Verify presence properly, signed out, on the actual losing queries. For each one, note whether an AI Overview appears and whether it expands to fill the page. Then note whether you're cited inside it, and who is when you're not. Rank trackers with SERP-feature detection can do this at scale.
The cited-or-absent question matters most, because ranking first and being cited in the Overview are independent outcomes. Each combination calls for a different play.
On tooling, practitioners note that Search Console added a limited generative-AI view under Performance in mid-2026. It shows impressions only, with no clicks and no query split, and it blends AI Overviews into AI Mode. Glance at it for direction. Then build the real split yourself with the intent and CTR method above, because the built-in view can't carry a diagnosis.
Check 5: overlay the calendar
Check the dates before you settle on a verdict. Google's Search Status Dashboard lists updates with start and completion timestamps, and the recent window is busy. A spam update ran August 18-21, and Overview expansion changed some informational SERPs in the same season. Volatility around updates is routine on top of that. If a decline starts inside an update window and concentrates on specific tactics or page types, it reads as algorithmic, even when impressions look stable.
One well-known search analyst has made a career out of this exact correction. He points out that sites crying "AI took our traffic" often show years of accumulating update damage in their visibility charts. It sounds harsh, but the charts usually agree with him.
Annotate your analytics with the dashboard dates and the model-release dates, permanently, and this check takes thirty seconds forever after. If your drop lines up with a core update, our guide to separating core update losses from AI search shifts walks you through the position test.
We'll show you exactly where your brand stands in AI search.
No commitment. $0 due today, cancel anytime. See how ChatGPT, Perplexity, Claude, Gemini, Grok, Google AI Mode, and Google AI Overviews talk about your brand today.
Check 6: rule out your own ship
Environmental suspects get the headlines, and self-inflicted ones get the actual traffic. Sites change themselves and then forget they did. So list anything deployed in the drop window, whether that was a migration or a redesign, a robots.txt edit or a CDN or bot-protection change, a consent flow or a URL restructure.
Cross-reference your deploy log against the drop date before you conclude anything about the outside world. Then ask whoever ships changes what else went out that week.
Check 7: check where the attention went
The last check reframes the loss. Pull your AI-referral traffic, meaning visits arriving from ChatGPT, Perplexity and friends, and add direct and brand-search trends over the same window. Some of the "lost" traffic simply moved channels. Buyers who used to click through informational results now meet you inside answers and arrive later, by name.
A brand whose informational clicks fall while its brand queries rise is being discovered differently rather than less. This is also the moment to check a scoreboard most sites don't have yet: whether the answers absorbing your clicks actually cite or mention you, which is what AI visibility measures.
Losing the click to an answer that recommends you is a very different event from losing it to an answer that recommends your competitor. Nothing in Search Console can tell you which one happened.
A worked hour, on a real shape
Run the case from that embedded thread through the checks and the verdict falls out on its own. Clicks fell roughly by half overnight, impressions stayed steady for three weeks, and every query was informational. The pages had already been appearing inside Overviews back when clicks were still growing.
Check 1 passes, since impressions holding means measurement and indexing are fine. Check 2 isolates cleanly to one intent band. Check 3 finds the fingerprint exactly, with stable presence and collapsing clicks.
Check 4 is where the interesting detail surfaces. The pages were already cited in Overviews while clicks were growing, so presence alone didn't cause the cliff. The plausible structural change is coverage and layout: more of the queries gaining Overviews at once, or existing ones expanding to fill the page. Both happened in waves this season.
The verdict is absorption with a layout shift as the accelerant, and nothing about the site is broken. The play follows directly. Keep the citations the pages already earn, since cited presence is the surviving asset on those queries, and shift new-content effort toward the intents that still click. Total diagnosis time on a real case was under an hour, most of it exports.
The client conversation
Agencies keep asking each other how to explain these drops to clients. The checklist hands you the script.
Open with what happened. The results page changed on the informational half of the client's queries, and the checks show rankings held while the answer layer absorbed the click. Then walk through what you verified, in order: tracking first, then the intent split and CTR at stable position. After that come Overview presence and the update calendar, and last, your own deploy log.
Next, say what changes now. You fight for citations inside the answers and reweight the calendar toward intents that click. You also stop measuring success purely in sessions on queries where sessions no longer exist. Finish with what the client will see weekly, which is presence in the answers, per engine, next to the traffic that remains.
That last line is the reporting shift this whole era demands. Clients accept it remarkably well when it arrives with evidence instead of excuses.
The verdict table
| What you found | The diagnosis | The play |
|---|---|---|
| Tags broken or ranges mismatched | Measurement, not market | Fix tracking, recheck |
| Position fell with clicks | Update or site issue | Audit against the calendar and policies |
| Stable position, informational CTR bleeding | Answer absorption | Fight for citations, reweight intents |
| Drop starts in an update window | Algorithmic | Policy and quality audit |
| Deploy log matches drop date | Self-inflicted | Roll back or repair |
| AI referrals and brand queries rising | Channel shift | Measure the answer layer, feed it |
After the diagnosis
If absorption is your verdict, recovery is a reallocation rather than a restoration. The informational click an answer consumed isn't coming back. The ground you can still win is being the source those answers cite, weighted toward the commercial intents where clicks survive.
You can measure that work weekly. Which engines cite you, and for which queries? Against which competitors, and how has each of those moved since the last check? RankControl runs exactly that scoreboard next to your rankings, on ChatGPT and Perplexity and also on Claude, Gemini, Grok and Google's AI surfaces. The next time a chart lurches, the first-hour question of absorption, update or accident gets answered by instruments you already have running.
The checklist above is the manual version. Either way, run it before you rewrite anything.
Know exactly what AI says about your competitors.
RankControl's Recon Agent monitors competitor citations across ChatGPT, Perplexity, Claude, Gemini, Grok, Google AI Mode, and Google AI Overviews. See where they show up and you don't.
