Keyword selection after AI Overviews: a 2026 framework for queries that still send clicks

Summary. Between 16 April and 15 July 2026, twelve answer-shaped queries earned this site 307,668 impressions and 909 clicks. That is a blended clickthrough rate of 0.295%. Over the same 90 days, three queries that were not answer-shaped earned 378 impressions and 22 clicks, a 5.82% CTR, roughly 20 times better on a fraction of the volume. One query ranked at average position 2.77 and converted 0.149% of its impressions, while another sitting worse at position 4.23 converted 16.07%. The position was not the variable. The shape of the question was. This matches the largest public study available: Ahrefs analysed 300,000 keywords and found that as of December 2025 the presence of an AI Overview correlates with a 58% lower CTR for the top-ranking page, up from 34.5% in their April 2025 study. Ranking first on a question Google answers itself is not a win. It is 168,145 impressions and 348 clicks, which is what "ios 27 release date" paid us. This article gives the scoring framework we now run every topic through before we write a word.

The uncomfortable part is that the failing keywords look like wins in every dashboard. High volume, good position, terrible economics.

The number that should end the position-one conversation

Ahrefs re-ran their AI Overview study using December 2025 data, comparing 150,000 keywords with an AI Overview present against 150,000 informational keywords without one, measured against a December 2023 baseline from before the rollout.

The findings are specific. In December 2023, average position-one CTR for informational keywords was 0.076. By December 2025 it had fallen to 0.039. For the keywords that now trigger an AI Overview, position-one CTR went from 0.073 to 0.016. Correcting for the general decline, the AI Overview itself accounts for roughly a 58% reduction.

Ryan Law, Director of Content Marketing at Ahrefs, put the arithmetic in a sentence:

"For every 100 clicks you could historically earn for a top-ranking page, Google now 'keeps' 58."
Ryan Law, Director of Content Marketing, Ahrefs

The per-position table from that study is the part most people miss, and it changes the strategy:

Position in search results	CTR impact when an AI Overview is present
1	-58.0%
2	-50.8%
3	-46.4%
4	-38.8%
5	-32.6%
6	-30.5%
7	-29.7%
8	-28.8%
9	-29.7%
10	-19.4%

Read that column downward. The damage is worst at position one and mildest at position ten. Climbing the rankings on an AI Overview query moves you into the zone where the AI Overview hurts most. You are not competing for the click. You are competing for what is left after the answer panel has taken it, and the better you rank, the larger the share it takes.

Ahrefs also found around 9% of AI Overviews appear outside position one, and their conclusion was blunt: AI Overviews "siphon away the majority of the clicks once available to top-ranking pages."

Corroboration is broad. Ahrefs cite Seer Interactive at 49.4% to 65.2%, Kevin Indig at over 50%, Authoritas at 47.5%, and the Daily Mail reporting 80% to 90% lower CTR. The range is wide. The direction is not in dispute.

What our own Search Console data shows

Public studies are averages across other people's sites. Here is what it looks like on one small site with a dominant topic cluster.

This is eCorpIT's Google Search Console data for 16 April to 15 July 2026. Our CTR by average position across the site:

Average position	Site CTR	Keywords at this position
1	28.44%	38
2	16.63%	37
3	10.82%	47
4	3.94%	79
5	3.56%	104
6	2.96%	128
7	1.14%	183
8	1.67%	148
9	2.90%	89
10	1.50%	62

That curve looks normal. Position one earns 28.44%, and there is a cliff between position 3 at 10.82% and position 4 at 3.94%, a 64% drop across a single rank.

Now put individual queries against that curve, and the picture breaks.

Query	Avg position	Impressions	Clicks	CTR	Site average at that position
ios 27	2.77	9,385	14	0.149%	10.82% at position 3
ios 27 release date 2026	4.60	7,163	6	0.084%	3.94% at position 4
ios 27 release date	8.44	168,145	348	0.207%	1.67% at position 8
when is ios 27 coming out	8.86	31,168	29	0.093%	2.90% at position 9
nextjs typescript 7	4.23	56	9	16.07%	3.94% at position 4

The comparison to sit with is the first row against the last. "ios 27" ranks at position 2.77 and converts 0.149% of impressions. "nextjs typescript 7" ranks worse, at position 4.23, and converts 16.07%. That is 108 times the CTR from a lower position.

In absolute terms it is starker. Fifty-six impressions produced nine clicks. Nine thousand three hundred and eighty-five impressions produced fourteen.

"ios 27" at position 2.77 runs about 73 times below our own site average for position 3. "ios 27 release date 2026" at position 4.60 runs about 47 times below our position-4 average. These are not ranking failures. Google is showing our page high up and answering the question above it.

The honest caveats

This is one site, and the sample is dominated by a single iOS 27 cluster, so the answer-shaped cohort is really one topic measured many ways. The non-answer-shaped cohort is small: 378 impressions across three queries, which is enough to be suggestive and not enough to be conclusive on its own. The site-average CTR figures above position 10 rest on tiny keyword counts and are noisy, which is why we have not used them.

What makes it worth publishing is that it points the same direction as a 300,000-keyword study, from the opposite end of the scale. The framework below is what we do about it.

Answer-shaped versus click-shaped

The distinction that predicts CTR is not head versus long tail, informational versus commercial, or volume versus difficulty. It is whether the query has a terminal answer.

A query is answer-shaped when a correct, complete response fits in a paragraph and nothing follows it. "ios 27 release date" is answer-shaped. The answer is a date. Once you have the date, there is nothing left to want. Any AI Overview that names the date has fully served the user, and your page is decoration below it.

A query is click-shaped when a correct response opens more questions than it closes, or when the value is in artefacts a summary cannot carry: a config file, a benchmark table, a decision under constraints, a number specific to the reader's situation.

Signal	Answer-shaped (avoid)	Click-shaped (target)
Query form	"release date", "what is X", "is X out yet", "when does X"	"X vs Y for Z", "how to do X with Y", "X cost breakdown", "why does X fail when Y"
Complete answer length	One sentence or one number	Multiple sections, tables or code
What the reader does next	Nothing, they have the fact	Applies it, compares, configures, budgets
AI Overview risk	Total, the panel is the product	Partial, the panel becomes a teaser
Our observed CTR	0.295% blended across 12 queries	5.82% blended across 3 queries
Volume	Usually high	Usually low
Value per impression	Near zero	High

The trap is that answer-shaped queries have the volume. "ios 27 release date" has 168,145 impressions in 90 days. "nextjs typescript 7" has 56. Every keyword tool will rank the first one higher on every metric except the only one that pays.

The click-survivability score

We score every candidate topic out of 10 before it enters the queue. Six questions, weighted by how well each predicted CTR in our own data.

1. Does a complete answer fit in one paragraph? If yes, score 0 on this dimension. If it needs sections, tables or code, score 3.
2. Can we produce something the panel cannot summarise? Original benchmarks, first-party data, a working config, a real price. Score 0 to 3.
3. Does answering it well require judgement under constraints? "Which should I use" beats "what is". Score 0 to 2.
4. Is the reader mid-task rather than mid-curiosity? Someone with a terminal open clicks. Someone idly wondering does not. Score 0 to 2.
5. Would a wrong answer cost the reader money or a rollback? Stakes drive verification, and verification drives clicks. Score 0 to 2.
6. Penalty: is the query a known AI Overview trigger shape? Subtract 2 for date, definition, yes/no or single-fact queries.

Anything scoring 6 or above gets written. Below 6, it does not, regardless of volume.

Applied to real candidates:

Candidate query	Q1	Q2	Q3	Q4	Q5	Penalty	Score	Verdict
ios 27 release date	0	0	0	0	0	-2	-2	Never
what is an AI Overview	0	1	0	0	0	-2	-1	Never
UCP vs ACP vs AP2 for merchants	3	3	2	2	2	0	12	Write
DPDP compliance cost for a startup	3	2	2	2	2	0	11	Write
ios 27 MDM migration for fleets	3	2	2	2	2	0	11	Write
is ios 27 out yet	0	0	0	0	0	-2	-2	Never

The scoring is deliberately harsh on question 1. In our data, one-paragraph-answerable queries never cleared 0.3% CTR regardless of position.

What to do with the answer-shaped rankings you already have

Most sites are not starting clean. We rank well on a cluster of queries that pay us almost nothing, which is an asset with the wrong monetisation, not a liability.

Three options, in descending order of how often they work.

Consolidate and re-point. If you have five pages competing on variants of one date query, you have five pages splitting a 0.2% CTR. Merge them into one timeline page and use it to link into the click-shaped work. The date query becomes a distribution channel for the article that actually earns, not a destination. Our GEO fixes for ranking first without clicks covers the consolidation mechanics.

Re-shape the page, not the keyword. You cannot make "ios 27 release date" click-shaped. You can make the page that ranks for it answer a second question the panel will not: what breaks in your MDM fleet on upgrade day. The query stays answer-shaped. The page stops being.

Accept it as brand impressions and stop measuring it as traffic. Three hundred thousand impressions at 0.3% is a weak traffic channel and a reasonable awareness channel. The mistake is booking it as the former and being confused by the revenue.

What does not work is chasing the ranking. We already rank 2.77 on "ios 27" and it converts 0.149%. Position one would not fix that. Per the Ahrefs table, position one is where the AI Overview takes the most.

Where the clicks went, and what still earns them

Zero-click search is not new, and AI Overviews are the latest in a sequence that includes Featured Snippets, Local Pack and Top Stories. Ahrefs make this point directly: CTR for non-AI-Overview informational keywords also fell over the same period, from 0.076 to 0.039. The panel accelerated a trend that predates it.

The strategic read is that the queries surviving are the ones where the value is not the fact. Four categories hold up in our data and in the published studies:

Comparisons with real benchmarks. Not "X vs Y" listicles, which summarise fine, but comparisons carrying numbers a panel cannot compress without losing the point.

Cost and pricing breakdowns. Specific, dated, with the assumptions visible. A summary of a cost model is not usable. The model is.

How-tos with code, configs or templates. The artefact is the product. A panel describing a config does not save anyone the work of writing it.

Decision-stage questions under constraints. "Which of these should we build on, given we are on Shopify and sell in India" has no terminal answer. It has a recommendation, and recommendations invite scrutiny, which is a click.

The common property is that a correct summary of the page increases rather than decreases the reader's need to open it. That is the actual test. If a perfect AI Overview of your article would satisfy the reader, do not write it.

The keyword sources that surface click-shaped queries

Search Console is a poor discovery tool for this work, because it only shows queries you already rank for, which skews toward what you have already written. It is a measurement tool, not a source.

Sources that surface click-shaped candidates:

Google Trends rising and breakout terms, filtered for anything with a task in it. Rising queries have not yet been saturated by content, and breakout terms often outrun the panel's confidence.

Product and event launch calendars, but targeting the second-order question. Not the release date. What the release breaks.

Competitor keyword gaps where the competitor's page is thin. A thin page ranking on a click-shaped query is an opening, because the query deserves depth and nobody has supplied it.

People Also Ask and community questions, read for the follow-up rather than the question. The first question is usually answer-shaped. The third one in the thread rarely is.

Your own support tickets and sales calls. These are pure click-shaped signal, because nobody opens a ticket about a release date.

India-specific considerations

Two things differ for teams targeting Indian queries.

Volume distributions are more extreme. The gap between a high-volume answer-shaped query and a low-volume click-shaped one is wider, which makes the score above harder to defend internally when someone points at the volume column. The arithmetic does not change. Twelve queries at 307,668 impressions paid us 909 clicks. Three queries at 378 impressions paid us 22. The second set cost far less to produce.

Query language mixing matters. Hinglish and transliterated queries are frequently answer-shaped in ways the panel handles well, while the genuinely click-shaped Indian queries tend to be pricing, regulation and vendor-selection questions where local constraints make the answer non-generic. DPDP, UPI and ONDC questions score well precisely because the correct answer depends on rules that are specific, changing and consequential. Our AEO, GEO and SEO guide covers how these surface differently across engines.

What we changed in our own pipeline

We used to queue topics by search volume and keyword difficulty. The iOS 27 cluster looked outstanding on both. It produced 307,668 impressions and 909 clicks in 90 days.

Three changes since:

Every candidate topic gets the survivability score before it is queued, and anything under 6 is dropped no matter what the volume says. We do not queue "release date" topics at all now.

Every article must contain at least one thing we produced rather than retrieved: a benchmark, a config, a table of our own numbers, a cost model. If we cannot name that artefact before writing, the topic is not ready.

We measure clicks per article, not impressions or position. The dashboards that made the iOS 27 cluster look like a success were measuring the wrong column.

The honest summary of our own data is that we spent 90 days ranking well for questions Google had already answered. The zero-click survival playbook and the GEO platform playbook cover what we built instead.

FAQ

What is click-survivability in keyword selection?

Click-survivability is whether a query still sends clicks once an AI Overview answers it. Queries with a terminal one-paragraph answer, like release dates or definitions, lose nearly all clicks. Queries needing tables, code, benchmarks or judgement under constraints keep them, because a summary increases rather than removes the reader's need to open the page.

How much do AI Overviews actually reduce clicks?

Ahrefs analysed 300,000 keywords and found the presence of an AI Overview correlates with a 58% lower CTR for the top-ranking page as of December 2025, up from 34.5% in April 2025. Corroborating studies range from Authoritas at 47.5% to Seer Interactive between 49.4% and 65.2%.

Does ranking first still matter with AI Overviews?

Less than it did, and not in the direction most teams assume. Ahrefs' per-position data shows the CTR damage is worst at position one at 58% and mildest at position ten at 19.4%. Climbing an AI Overview query moves you into the zone where the panel takes the largest share of available clicks.

Which keywords still send clicks in 2026?

Four types hold up: comparisons carrying real benchmarks, cost and pricing breakdowns with visible assumptions, how-tos shipping code or configs, and decision-stage questions constrained by the reader's situation. The shared property is that a correct summary of the page leaves the reader needing the page.

How do I test whether a keyword is answer-shaped?

Write the complete correct answer. If it fits in one paragraph and nothing follows it, the query is answer-shaped and an AI Overview will serve it fully. If the answer needs sections, a table, a configuration or a recommendation the reader might argue with, it is click-shaped and worth writing.

Should I delete pages ranking on answer-shaped queries?

Usually not. Consolidate variants into one page, then use that page to route readers into work that earns clicks. Treat the impressions as an awareness channel rather than a traffic channel. Deleting discards distribution; the error is booking those impressions as traffic and being surprised by the revenue.

Why did a page at position 4 outperform one at position 2.8?

Because shape beat position. In our 16 April to 15 July 2026 Search Console data, "nextjs typescript 7" at position 4.23 converted 16.07% of impressions while "ios 27" at position 2.77 converted 0.149%. The first query implies a task with no terminal answer. The second is a fact Google states directly.

Is Search Console enough for keyword research now?

No. Search Console only reports queries you already rank for, so it reflects what you have written rather than what is worth writing. Use it to measure click-survivability after publishing, and use Google Trends rising terms, People Also Ask follow-ups, competitor gaps and your own support tickets to find candidates.

How eCorpIT can help

eCorpIT is a CMMI Level 5 certified, senior-led engineering organisation in Gurugram that runs this framework on its own content before recommending it to anyone. We audit which of your rankings are earning clicks and which are decorating an answer panel, score your topic pipeline for click-survivability, and rebuild content around artefacts a summary cannot replace. Talk to our team via /contact-us/ about a search visibility and GEO review.

References

1. Update: AI Overviews Reduce Clicks by 58%. Ryan Law and Xibeijia Guan, Ahrefs, 4 February 2026
2. AI Overviews Reduce Clicks by 34.5%. Ahrefs, April 2025
3. Welcome to Zero-Click Search. Ahrefs
4. How SERP Features Have Evolved in the AI Era. Ahrefs
5. What Triggers AI Overviews? 86 Factors and 146 Million SERPs Analyzed. Ahrefs
6. How to Rank in AI Overviews: What Actually Works. Ahrefs
7. AI Overviews Change Every 2 Days. Ahrefs
8. How to Track AI Overviews: Mentions, Citations, Click Loss. Ahrefs
9. Google zero-click searches reach 68% in early 2026: Study. Search Engine Land
10. GSC data is 75% incomplete. Kevin Indig, Growth Memo
11. Google AI Overviews: publishers report clickthrough declines. Press Gazette, on the Authoritas report
12. Google AI Overviews Reduce Clicks By 58%, Study Finds. Medianama, February 2026
13. Google AI Overviews statistics: key CTR and traffic insights. Psyke
14. eCorpIT Google Search Console, property ecorpit.com, 16 April to 15 July 2026. CTR by average position and query-level clicks, impressions and position. First-party data reported in this article.

Last updated: 16 July 2026.