Reframe Keyword Research for Zero-Click: Metrics That Matter
Clicks are drying up as AI Overviews answer in place. Reframe keyword research for zero-click with four citation-likelihood metrics that predict citations.
Your rank-one page can now lose the click before anyone sees it. Google's AI Overviews answer the query at the top of the results, and the traditional blue link sits below the fold, unread. If your keyword research still scores opportunities on volume and estimated clicks, you are optimizing for a funnel that is quietly closing. This piece reframes keyword research for zero-click search around one question: when an AI Overview fires for a query, how often does your page get cited? Track that, and you can rebuild your keyword model for the search page we actually have in 2026 — not the one we had two years ago.
How Google AI Overviews Are Rewriting the Value of a Click
The click math already moved. When an AI summary appears, users click a traditional search result in just 8% of visits, versus 15% when no summary shows, per Pew Research Center browsing data. Roughly half the clickthrough evaporates the moment an Overview renders. Ahrefs found the same pull at scale: across 300,000 keywords, an Overview cut the position-one clickthrough rate for informational queries by about 34.5% — Ahrefs' 2025 clickthrough study.
And Overviews are not a niche SERP feature. They fire on 53% of searches ten words or longer, and on 60% of queries that begin with a question word like who, what, or why, per the same Pew analysis. That is exactly the informational, long-tail territory most content programs target. The search volume did not disappear; the click did. So a KPI built on clicks now understates the value of a page that is being read — as a citation — every day without ever earning a visit. If you keep grading keywords on clicks alone, you will retire your best zero-click assets for looking like dead weight in the dashboard.
The Citation-Likelihood Metrics That Replace Click KPIs
So measure the thing that now creates value: citation. I score every target query on four signals I can pull or compute without waiting a quarter for rankings to settle. Think of them as the citation-likelihood stack — Impression-to-Citation Rate, Direct-Answer Share, Factual Density, and Provenance. No single one is enough on its own; together they tell you which pages are already earning AI trust and which keywords are worth chasing under the new rules. That is the same discipline this site's guide to revaluing keywords for AI SERP summaries applies at the keyword-selection stage, before you ever start measuring citations.
Impression-to-Citation Rate (ICR): Definition and Formula
ICR is the headline metric. Definition: of all the AI-Overview impressions your page is eligible for on a query, what share cite it as a source? The formula is ICR = overview citations of your URL / overview impressions for the query. If a query fires an Overview on nearly every search and your page is named in the source list four times out of every ten checks, your ICR is 0.4. Pull the denominator from a daily SERP check on your tracked queries; pull the numerator from the cited-source list inside the Overview.
As a working baseline, treat brand-new pages as near-zero, and expect strong, well-structured answers on mid-competition queries to land in the 0.2 to 0.4 range once indexed. Calibrate those bands against your own log within a few weeks — the absolute numbers matter far less than the trend line per page. Any page holding above 0.5 is a template worth cloning across the cluster.
Direct-Answer Share: Who Owns the Canonical Answer
ICR tells you whether you are cited. Direct-Answer Share tells you whether you own the answer. For a keyword cluster, it is the fraction of Overviews across that cluster that pull their canonical sentence — the definition, the number, the step — from your domain rather than a competitor's. A Direct-Answer Share of 0.4 means your phrasing is the one the model reaches for in four of every ten answers in that cluster. This is the metric that most resembles old-school rank tracking, except the unit is whose words got quoted, not whose link sat at position one. Owning the canonical phrasing beats ranking, because the Overview can quote you while sending the click nowhere — and the quote still builds the brand recognition that drives branded search later.
Factual Density Score: Making a Soft Signal Countable
Models favor content that is dense with checkable facts, and "quality" is useless as a metric until you can score it. So make it countable. My Factual Density rubric is three numbers per page: verifiable facts per 100 words, whether a concise one-to-two-sentence answer sits in the first screen (yes or no), and the count of primary-source links. Score each target page, rank the corpus, and you have a rewrite queue ordered by citation-readiness instead of gut feel. Google's own guidance on people-first content leans the same way — it tells raters to look for clear sourcing and demonstrable expertise, per Google Search Central. Density is not stuffing numbers in; it is swapping vague claims for attributed ones.
Provenance Signals: The Trust Markers Models Prefer
Provenance is the trust surface around the content: a visible author with real expertise, a dated last update, inline citations to primary sources, structured data that matches the page, and a domain with topical authority. One caution that saves wasted effort — Google states plainly that there is no special schema and no new markup that gets you into AI Overviews, and that your structured data must match the visible text, per the AI features guidance. Provenance is not a schema hack. It is the same E-E-A-T surface Google already rewards, now pulling double duty as a citation signal. Track it as a simple checklist score per page, then watch whether pages that clear the full checklist earn higher ICR over your first few weeks of logging.
How to Instrument These Metrics in Days, Not Months
You do not need a data warehouse. You need a daily snapshot and somewhere to log it. The fast build is four steps.
- Export your tracked queries from Search Console, starting with impression-heavy, click-light queries — a widening gap between impressions and clicks is the fingerprint of an Overview eating your traffic. If yours is already bleeding, pair this with an organic-traffic recovery checklist so you are fixing, not just measuring.
- Run a daily SERP check on those queries — a scriptable SERP API, or Screaming Frog custom extraction against a rendered results page — and record two things: did an Overview fire, and which URLs did it cite?
- Write each check to a flat citation log: date, query, overview-fired, your-URL-cited, competitor-URLs-cited. A single sheet or small table is enough to start.
- Compute ICR and Direct-Answer Share weekly from the log, and score Factual Density and Provenance by hand on the pages that matter.
Two gotchas from running this. First, Overviews are volatile — the same query can fire an Overview one day and not the next, so a single scrape lies and only the rolling weekly rate is trustworthy. Second, tag any experiment traffic with UTMs so you can still see whether the citations that do click are worth more or less than the old organic clicks. Instrument first, theorize later.
Fast Playbook: Reframe Keyword Research for Zero-Click Citations
Now put the metrics to work. This is the loop I run.
- Re-score the keyword list. Add an "Overview-fires" column and an ICR column next to volume, and stop taking third-party keyword tools at face value. A mid-volume query where an Overview always fires and your ICR is climbing outranks a high-volume query where you sit at position one but the click is already gone.
- Rewrite the highest-impression, lowest-ICR pages first. Lead with a concise 40-to-60-word answer to the exact question, in the first screen, before any preamble. That block is what gets quoted.
- Back the answer with an evidence row — a stat, a source link, a date — inside the first two paragraphs, then structure the rest as scannable, self-contained chunks a model can lift cleanly.
- Design the change as an experiment: edit one page, hold the rest, and read weekly ICR and Direct-Answer Share for four to six weeks before judging. Set a target — say, move ICR from near-zero to above 0.3 — and a rollback rule before you start.
On rollback: if being cited coincides with a real drop in downstream signups or revenue — cited but bypassed — stop chasing the citation. De-optimize that page back toward the click, keep the answer thinner, route the query to a channel you control, and keep every target tied to a real conversion rather than a vanity ranking. Citation is the new default KPI, not a religion. The money still has to close somewhere, and the whole point of measuring is to catch the page where visibility and revenue have quietly split. Converting that visibility into revenue instead of a vanity metric is worth its own playbook — see how to convert AI overviews into subscribers and revenue.
How VarynForge Fits In
VarynForge scores keyword opportunities the way this article argues you should — modeling citation-likelihood alongside volume, so your content plan is tuned for AI Overviews rather than clicks, and exporting a prioritized brief queue you can start rewriting the same day. If you would rather not hand-build the entire scrape-and-score pipeline before you learn anything, that is the shortcut: VarynForge keyword research.
Further Reading
- LLM Intent Classification: A RAG Pipeline for Keyword Research
- Keyword Strategy for SEO: Templates and Prioritization Playbook
- Types of Search Intent: A Vector Framework for Content Teams
- Keyword Research for SEO: From Seed Terms to High-Intent Targets
Sources
- Pew Research Center — Google users are less likely to click on links when an AI summary appears (2025)
- Ahrefs — AI Overviews Reduce Clicks by 34.5% (2025)
- Google Search Central — AI features and your website
- Google Search Central — Creating helpful, reliable, people-first content
- Google Search Central — Introduction to structured data markup
Key Takeaways
Clicks are no longer the value they were; citations are. Reframe keyword research for zero-click by scoring queries on Impression-to-Citation Rate and Direct-Answer Share, make quality countable with a Factual Density rubric, and treat provenance as the E-E-A-T surface Google already rewards rather than a markup trick. Stand the instrument up in days — Search Console, a daily SERP scrape, a citation log — rewrite your highest-impression, lowest-ICR pages to lead with the quotable answer, and test for a few weeks before you trust any single number. The team that measures citation first will reprice its whole keyword list while the teams still counting clicks are still wondering where the traffic went.
Frequently asked questions
What exactly are Google AI Overviews and how do they affect organic clickthrough rates?
AI Overviews are the AI-generated summaries Google places at the top of many results pages, answering the query directly and citing a handful of source URLs. Because the answer sits above the traditional links, fewer people scroll down to click. Pew Research found users click a normal result far less often when a summary appears, and Ahrefs measured a sizable drop in position-one clickthrough on informational queries. The practical effect is that a page can rank first, be read as a citation inside the Overview, and still lose the visit. That is why clickthrough alone now understates a page's real value.
How do I calculate Impression-to-Citation Rate (ICR) for a page or query?
ICR is the share of AI-Overview impressions on a query where your URL is named as a source. The formula is ICR = overview citations of your URL divided by overview impressions for the query. Get the denominator by running a daily SERP check on your tracked queries and counting how often an Overview fires. Get the numerator by recording, on those same checks, how often your URL appears in the Overview's cited-source list. Compute it weekly rather than daily, because Overviews are volatile and a single scrape is noisy. Watch the per-page trend line more than the absolute number.
Can structured data or schema markup increase the chances my content is cited in an AI Overview?
Not on its own. Google states there is no special schema and no new markup that gets a page into AI Overviews, and that any structured data you use must match the visible text on the page. Schema still helps Google understand and correctly represent your content, so it is worth keeping accurate, but treating it as a citation cheat code is wasted effort. The signals that actually move citation likelihood are a clear, quotable answer near the top of the page, dense verifiable facts, primary-source links, and genuine author and site authority. Fix those first, then keep your markup honest and current.
What quick experiments can I run to test whether a page becomes more likely to be cited?
Pick one high-impression, low-ICR page and change a single variable: add a concise forty-to-sixty-word answer to the exact question in the first screen, above any preamble. Hold every other page steady so the test stays clean. Then read weekly ICR and Direct-Answer Share for that query cluster over four to six weeks before you judge the result. Set a target in advance, such as moving ICR from near-zero to above 0.3, plus a rollback rule. Single-page, single-variable tests are slower than bulk edits but they actually tell you which change earned the citation.
If my traffic drops because of AI Overviews, how can I measure whether being cited still delivers value?
Separate the two things that used to travel together: visibility and revenue. Keep tracking Direct-Answer Share so you know your brand is still the quoted source, and tag any clicks that do come through with UTMs so you can compare their downstream conversion to your old organic clicks. If a page is heavily cited but its signups or sales fall, visibility and revenue have split, and the citation is not paying you. If cited pages still convert the visits they send, being quoted is building branded demand you can measure later. Instrument both sides before drawing conclusions.
Will trying to optimize for AI Overviews risk a Google penalty for gaming the system?
Not if you optimize for answer quality rather than tricks. Google's guidance is explicit that no special optimization is required for AI features and that ordinary SEO fundamentals still apply: crawlable pages, content in real text, structured data that matches what readers see, and helpful, people-first writing. Leading with a clear answer, citing primary sources, and showing real expertise are exactly what Google already rewards, so there is nothing to penalize. The risk comes from manipulation, such as schema that misrepresents the page or thin content dressed up with fake authority signals. Build for the reader and the citation follows safely.
