Trends & Insights Report
Why rankings and AI citations need separate measurement
Ahrefs reported 76% and 38% overlap among already-cited AIO URLs in two separately published studies. Their samples and citation parsing differ; rankings and citations need separate measurement.
v1.1 · October 2026 · Tenten GEO Research
CH.01
What the overlap studies measure
Ahrefs reported 76% top-ten overlap among already-cited AIO URLs in July 2025. Its March 2026 study of 863K keyword SERPs and 4M AIO URLs reported about 38% within the first ten SERP blocks. The samples and citation parsing differ.
These percentages start with URLs already cited in AI Overviews and ask where they rank. They do not tell us a top-ten page's probability of being cited. The two studies do not establish a same-page, same-method time series.
| Timeframe | Share of cited URLs in top-ten results | Source |
|---|---|---|
| 2025-07-21 | 76% | Ahrefs, July 21, 2025 (1.9M citations from 1M AIOs; top three visible citations) |
| 2026-03-02 | 38% | Ahrefs, March 2, 2026 (863K SERPs / 4M AIO URLs; revised citation parsing; SERP blocks) |
Reported overlap between top-ten rankings and AI citations; see the chart note and methodology for differences in methods and measurement baselines.
Top-ten overlap among AIO citations: separate published observations
Ahrefs, July 21, 2025 and March 2, 2026; separate samples and methods
The 76% and 38% figures come from separate Ahrefs studies with different samples and parsing. The rounded 38% uses SERP blocks including features; organic-only overlap was 37.1%. These are shares of already-cited URLs, not citation probabilities for ranking pages. This chart is free to share: please cite the source: Tenten GEO, The Rank-Citation Decoupling Report.
Chapter takeaway
Track rankings and citations separately. These overlap observations do not establish a fixed-cohort decline or a ranking-proxy failure rate.
CH.02
Related queries can surface different sources
Google says AI Overviews and AI Mode may run related searches through query fan-out. A page can therefore appear as a source even when it is outside the top ten for the original query.
This does not divide search into a page-only system and a passage-only system. Google Search already uses passage ranking to understand sections within a page. Clear explanations help readers and make the information easier to interpret; they do not create a citation guarantee.
A difference between ranking and citation is an observation to investigate. It does not identify a single cause. Record the question, engine, answer date and cited URLs before deciding whether relevance, access, competing sources or changes in the answer deserve attention.
| Dimension | Search result observation | AI answer observation |
|---|---|---|
| Question | Record the exact search query | Record the prompt and conversation context |
| Result | Record the URL and its position | Record the answer and its cited URLs |
| Scope | Fix location, device and date | Fix engine, mode, language and date |
| Related topics | Check relevant queries separately | Treat proposed fan-out questions as hypotheses |
| Decision | Investigate changes in search performance | Investigate changes in answer inclusion |
A measurement checklist, not a description of every engine’s internal pipeline.
Chapter takeaway
Measure the two surfaces directly. Neither a ranking screenshot nor one AI answer establishes how consistently a brand is visible.
CH.03
A domain score cannot replace observed citations
Domain metrics and citation outcomes measure different things. A correlation must be read with its engine, sample, metric definition and uncertainty; it does not establish the effect of changing a score.
Ahrefs’ June 2025 study compared mention share with Ahrefs Rank for the 50 most-mentioned domains in each of three systems. The reported relationships varied by engine, and the author cautioned that this restricted sample may not represent other domains. It is not a universal domain-authority prediction rule. https://ahrefs.com/blog/domain-link-metrics-not-good-mention-predictors/
Keep useful link and ranking measures in the SEO report, and record citations separately. Before reallocating budget, check whether the proposed change addresses a demonstrated access, relevance or evidence gap on the page.
Chapter takeaway
A weak association in one sample does not establish that links are useless, or that rewriting a paragraph will cause more citations.
CH.04
Four checks from an editorial evidence review
Cyrus Shepard’s May 7, 2026 review selected 54 studies, experiments, explainers and patents. He assigned scores by judgment, considering repeated findings, evidence strength and official support. The values below are his assessments, not engine weights or measured effect sizes.
Selected author-assigned evidence scores
Cyrus Shepard / Zyppy, May 7, 2026; editorial scoring
The scores do not define four mandatory conditions or an expected citation lift. They summarize the author’s reading of varied sources. https://signal.zyppy.com/p/ai-citation-ranking-factors
Use this review to identify questions worth checking on your own site. Confirm actual indexing, retrieval and answer behavior for the target system; do not treat a score of 9.4 as a requirement to rank in the top ten.
| Factor | Author score | Plain-English meaning | What you do about it |
|---|---|---|---|
| URL accessibility | 9.5 | Whether intended crawlers can access the URL | Check robots, HTTP responses and rendered content for the intended search system; keep training policy separate |
| Search ranking | 9.4 | How the page ranks for the original query | Maintain useful SEO work and record the actual result |
| Fan-out ranking | 9.3 | How sources appear for related queries | Cover relevant buyer questions without pretending to know hidden fan-out queries |
| Preview control | 9.2 | Directives limiting previews or text use | Review nosnippet, data-nosnippet and max-snippet against the intended publishing policy |
Selected checks; confirm platform-specific controls in the provider’s documentation.
Chapter takeaway
Preview controls are technical directives. Clear headings and complete answers are useful content practices, but they are a different issue.
CH.05
Choose the next change from the observed problem
Start with access and eligibility, then investigate relevance and content. Keep the change small enough to evaluate against a recorded baseline.
For Google AI Overviews and AI Mode supporting links, the page must be indexed and eligible for a Search snippet. Google does not require a top-ten position or a special AI file. Meeting the requirements still does not guarantee inclusion.
STEP 1
Check access for the intended system
Inspect the actual response and readable content. Separate search crawler access from training permissions; blocking a training bot alone does not establish that search retrieval is blocked.
STEP 2
Check search eligibility and relevance
For Google, inspect indexing and snippet eligibility. Compare the page with the buyer’s question; use ranking as one observation, not a citation threshold.
STEP 3
Check preview controls and answer clarity separately
Confirm that preview restrictions match your policy. Then make the answer complete, specific and easy for a reader to verify. Do not remove intentional restrictions just to pursue visibility.
STEP 4
Investigate related buyer questions
Use product knowledge, search evidence and customer questions to choose relevant subtopics. Proposed query expansions are hypotheses until observed.
STEP 5
Measure the change on both surfaces
Keep the prompt set and evaluation conditions stable. Compare ranking and answer citations before and after the change, and record other changes that could explain the result.
Chapter takeaway
Prioritize a demonstrated blocker or missing answer. The evidence does not support a universal rule to stop improving rankings once a page reaches the top ten.
CH.06
Track rankings and citations with explicit denominators
Keep rankings and citations side by side, with the query or prompt, engine, date and sample size. An observed gap tells you where to investigate; it does not explain why it happened.
If ranking stays stable while citations fall, check repeated answers and source changes before rewriting. Model changes, query context, sampling variation and competing sources can all complicate a before-and-after comparison.
| Quadrant | Rank | Citation rate | Reading | Action |
|---|---|---|---|---|
| Both observed | Top ten | Cited in the sample | The page appeared on both surfaces | Retain the evidence and monitor under the same conditions |
| Citation gap | Top ten | Rarely cited in the sample | Search visibility did not imply answer inclusion | Check access, question relevance and the sources actually cited |
| Search gap | Outside top ten | Sometimes cited | A citation does not require an original-query top-ten position | Investigate useful SEO improvements without assuming a content defect |
| Neither observed | Outside top ten | Not cited in the sample | Neither result was observed within this measurement | Check eligibility, sample coverage and the value of the page |
A diagnostic reading of the observed sample, not causal attribution.
Weekly
Citation rate (fixed prompt set × multiple engines)
Use a fixed prompt set across the selected engines. Define citation rate as eligible answers containing the target URL divided by all eligible answers observed, and report the count alongside it. Record errors and missing answers separately.
Monthly
Rank × citation gap review
Review the pages with an observed gap and inspect the underlying answers. Distinguish one-off variation from repeated changes before assigning work.
Quarterly
Resource allocation decision
Use search performance, citation observations and verified business outcomes together when setting priorities. Citation rate alone does not establish qualified demand or revenue.
What to retain in the report
Keep the question, answer, cited URL, date and denominator. These records let the next reviewer distinguish an observed change from an explanation that still needs evidence.
Sources & methodology
Sources checked October 5, 2026: Ahrefs, July 21, 2025 and March 2, 2026, report separate samples of already-cited URLs; the 76% and 38% figures must not be treated as a fixed-cohort time series or a ranking-page citation probability. https://ahrefs.com/blog/search-rankings-ai-citations/ https://ahrefs.com/blog/ai-overview-citations-top-10/ The domain-metric discussion uses the limited top-50 sample described by Ahrefs on June 25, 2025. https://ahrefs.com/blog/domain-link-metrics-not-good-mention-predictors/ Zyppy’s May 7, 2026 scores are an author-assigned evidence review, not confirmed ranking weights or a statistical meta-analysis. https://signal.zyppy.com/p/ai-citation-ranking-factors Technical eligibility and preview controls follow Google’s AI features documentation; Google’s ranking systems guide also describes passage ranking. https://developers.google.com/search/docs/appearance/ai-features https://developers.google.com/search/docs/appearance/ranking-systems-guide The workflow and diagnostic table are Tenten’s recommendations, not measured client outcomes. Original studies have different populations and methods; no combined effect or causal lift is calculated.
© 2026 Tenten GEO · Feel free to cite this whitepaper, please credit the source.
Discuss your AI search priorities
Request a 30-minute benchmark interpretationSend a booking request. We’ll email you to confirm a time within 24 hours.