Short answer, as of August 2026: if your rankings and impressions held steady but clicks fell, the most likely cause is a generated answer sitting above your result and satisfying the query before the user scrolls. The fix is not a ranking fix. It's making your page the source that answer cites, and moving your commercial value to the queries a generated summary can't resolve.
What actually changed
Google now composes answers at the top of many results pages. AI Overviews is the summary block above the organic listings, assembled from multiple sources with links out to them. AI Mode is a fuller conversational surface where the response is the result page, with a supporting set of cited links.
Two consequences follow, and only one is widely understood.
The obvious one: a query answered in the summary produces fewer clicks. Definitional and informational queries lose the most, because they were always one-fact queries that a snippet could satisfy.
The one that matters more: the surface has become a citation surface rather than purely a ranking surface. Being named and linked inside the generated answer is now a distinct thing to compete for, and it does not follow automatically from ranking first. Pages ranked below the fold get cited; pages ranked first sometimes don't.
Reading your Search Console data correctly
The signature of this shift is specific, and misreading it sends people chasing the wrong fix.
Impressions flat or up, clicks down, average position stable. That's a click-through-rate change caused by the surface, not a demotion. An impression is still recorded when your listing appears below a generated answer — you're visible, just less clicked.
A split by query type. Informational and definitional queries drop hardest. Transactional, navigational and branded queries hold up, because a summary can't complete a purchase, log you in, or replace a specific company. If your decline is evenly spread across all query types, look for a different cause.
| Query type | Example | Click exposure | What the page is now for |
|---|---|---|---|
| Definitional | "what is X" | High | Citation value; stop expecting traffic |
| Informational how-to | "how do I fix X" | High | Citation value plus brand exposure |
| Cost / pricing research | "how much does X cost" | Medium | Cited and clicked when specifics are yours |
| Comparison | "X vs Y for Z" | Medium | Being named in the shortlist is the win |
| Local intent | "X near me" | Low–medium | Entity accuracy matters more than copy |
| Transactional | "book X", "buy X" | Low | Unchanged — protect and prioritise |
| Branded / navigational | "[your company] reviews" | Low | Unchanged — but audit what's said about you |
No isolating dimension exists. Search Console does not separate clicks originating from AI Overviews or AI Mode from ordinary organic clicks, and neither does GA4 — they arrive as google / organic either way. You are inferring from the pattern, not measuring directly. That measurement gap, and what to pair it with, is covered in why AI citations barely show up in analytics.
Before concluding anything, rule out the ordinary explanations: seasonality against the same period last year, a core update, a site migration, a tracking change, or a genuine ranking loss. Check average position per query. If position is unchanged and CTR fell, the surface is the story.
What to do about it
Four responses, in the order they pay off.
1. Make your page citable, not just rankable
Generated answers pull specific, verifiable statements — a figure, a threshold, a named exception, a direct definition. Prose that circles a topic for four paragraphs before saying anything concrete provides nothing to lift.
The research here is unusually clear. The Princeton-led GEO study tested content modifications against generative engines and found that adding citations to authoritative sources, statistics and direct quotations lifted visibility in generated responses by up to 40%, while keyword stuffing — the reflex from a decade of SEO — produced no benefit at all.
Practically: answer the question in the first two sentences of the relevant section, state figures with units and dates, name the exceptions, attribute claims to real sources. The detailed standard is in how to get cited by ChatGPT and Perplexity, and how to check a draft against it before publishing is in scoring a draft for AI content optimization.
2. Rebalance toward queries a summary can't finish
Audit your top pages by what the query wants. "What is X" is now largely a summary's job and its click value is structurally lower — keep the page for citation value, stop expecting traffic from it. Shift commercial emphasis to queries that require your specific offer: pricing for a defined situation, availability, comparisons involving your product, anything transactional.
This is a portfolio rebalance, not an abandonment. Informational pages still earn citations, and citations still produce brand exposure and branded search. They just stop being a traffic line item.
3. Fix the mechanical layer
A generated answer can only cite what it can fetch and read. If your WAF blocks crawlers, or your key content only exists after JavaScript runs, you're ineligible for the citation regardless of content quality. Those two checks are in the crawler access audit and why client-side rendered content is invisible to AI engines. Google renders JavaScript, so a client-rendered site is less exposed here than with other engines — but it remains fully exposed everywhere else.
4. Track the other engines too
Optimising only for Google's surfaces means measuring one of six places your buyers now ask questions. ChatGPT, Claude, Perplexity, Gemini and Copilot each weight sources differently, and being cited well in one while absent from another is common and diagnosable — the per-engine differences are in how the major engines pick their sources.
What not to do
Don't chase a "rank in AI Overviews" hack. There is no placement to buy or setting to toggle. Inclusion follows from being a retrievable, specific, corroborated source on the question.
Don't block Google's AI features to protect clicks. Google-Extended governs Gemini training and grounding use, not AI Overviews eligibility, and content that ranks in Search can appear in the summary. Blocking Googlebot to opt out removes you from Search entirely — a cure considerably worse than the disease.
Don't delete informational pages that lost traffic. They're doing citation work now. Removing them removes your source claim on the topic and typically loses ranking for adjacent queries too. Retrofit them rather than pruning them.
Don't panic-rewrite everything. Start with the pages that lost the most clicks and already rank in the top ten. Retrofitting a page that's already retrievable is far higher-return than earning retrieval from scratch on a new one.
The uncomfortable strategic point
Some traffic is not coming back. A query that used to send a click and now gets answered in place is a structural change, not a downturn you can optimise through.
What replaces it is a different currency: being the source the answer names. That produces branded search, direct visits and calls rather than trackable clicks — real demand that your click-based reporting is structurally poor at seeing. Teams that keep score purely on organic sessions will conclude the channel is dying while their branded search volume quietly climbs.
The budget consequence — how much to move from conventional SEO into this work, and how fast — is worked through in splitting a budget between GEO and traditional SEO.
Getting a baseline
The practical starting point is knowing where you stand before you change anything. Define the questions your buyers ask, sample them across the engines, and record how often you're named — including Google's AI Mode alongside the standalone assistants.
DigiRank's AI Visibility Tracker samples ChatGPT, Claude, Gemini, Perplexity, Grok and Google AI Mode on a schedule and charts share of voice per prompt over time, with an opportunities view listing the prompts you're missing from. Because it reads Search Console and GA4 in the same tool, you can put the CTR decline on informational queries and the AI-side citation trend on one timeline instead of arguing about them separately. It starts at $99/mo on Starter, with the Agency tier at $249/mo covering up to 15 client tenants on a 14-day trial. How to choose the prompts is in building a prompt set worth tracking.
Frequently asked questions
Why are my impressions steady but my clicks dropping? That pattern usually means a generated answer is appearing above your result and satisfying the query before the user scrolls. Your listing still registers an impression, so visibility looks unchanged while click-through falls. Confirm by checking average position — if position held and CTR fell, the surface changed, not your ranking.
How do I rank in AI search results like ChatGPT and Perplexity? There's no placement to buy. Inclusion requires three things in order: the engine's crawler must be able to fetch your page, the content must be in the raw HTML, and the page must contain specific verifiable statements worth quoting — figures with dates, named exceptions, sourced claims. Research on generative engines found citations, statistics and quotations lifted visibility by up to 40%, while keyword stuffing did nothing.
Can I opt out of AI Overviews without losing Google traffic?
Not meaningfully. Google-Extended controls Gemini training and grounding use rather than AI Overviews eligibility, and content eligible for Search can appear in the summary. Blocking Googlebot would remove you from Search altogether, which costs far more than the clicks the summary absorbs.
Which queries lose the most traffic to AI Overviews? Informational and definitional queries lose most, because a summary can fully answer them. Transactional, navigational and branded queries hold up much better — a generated answer can't complete a purchase, log a user in, or substitute for a specific company they're looking for.
Should I delete informational pages that stopped getting clicks? No. Those pages are now doing citation work — being the source a generated answer names — which produces brand exposure and branded search rather than direct clicks. Deleting them removes your claim on the topic and usually costs ranking on adjacent queries. Retrofit them to a citation standard instead.
Can I see in Search Console which clicks came from AI Overviews?
No. Search Console does not separate AI Overviews or AI Mode clicks from ordinary organic clicks, and GA4 records both as google / organic. You infer the effect from the pattern — flat impressions with falling CTR concentrated on informational queries — rather than measuring it directly.
Is optimizing for AI Overviews different from optimizing for ChatGPT? The content standard is largely the same: specific, sourced, directly-answering material. The mechanics differ. Google renders JavaScript and uses its own index, while most standalone assistants read raw HTML and draw on different indexes, so a site can be cited well in one and absent from another. Track all of them rather than assuming one result generalises.