Generative engine optimisation — tactical patterns for 2026

Generative engine optimisation (GEO) is the operational practice of designing pages so they are surfaced and cited inside the answers produced by large-language-model search systems — Google AI Overviews, Google AI Mode, ChatGPT search, Perplexity, Microsoft Copilot — rather than (or in addition to) ranking inside the traditional ten blue links. It is a sibling discipline to search engine optimisation: the underlying crawl and indexing pipeline still matters, but the optimisation target moves from a position on a results page to a citation slot inside a synthesised answer. The tactical patterns documented here describe what the published evidence — peer-reviewed, independent, and vendor — says actually moves citation rates in 2026, what the supporting prevalence and behavioural data look like, and which patterns the same evidence base shows do not work.

This page is the practitioner-level companion to AI Overview citation patterns (GEO/AEO), which covers the foundational published evidence on citation rates and remains the canonical reference for the underlying studies. The cross-cutting editorial rules that govern how sources are cited inside Candid material are kept in Editorial discipline and sourcing. Discovery and indexing fundamentals — the layer underneath GEO — are documented at How Google crawls, discovers, and indexes pages. Schema and the technical-SEO surface area sit at Technical SEO standards and structured-data discipline (2026).

The 2026 LLM-answer-engine landscape

The current generative-search landscape is a multi-surface ecosystem rather than a single replacement for Google's classic SERP. Each surface draws on a different index, with different citation behaviour, and the launch dates anchor any discussion of "when AI search arrived."

Search Generative Experience was announced at Google I/O on May 10, 2023 as an opt-in Labs feature, and rebranded and launched as AI Overviews in the US on May 14, 2024, on by default for some queries.

Source: Google I/O 2023 keynote; Google blog announcement May 14, 2024. Confidence: Verified.

AI Overviews reached Canada on the week of October 28, 2024, after a small-percentage test and phased "over the coming weeks," in English, Hindi, Indonesian, Japanese, Portuguese and Spanish. French was explicitly not supported at launch — a material caveat for Quebec-market work.

Quote (Google Canada blog): "Starting this week, we are beginning the full rollout of AI Overviews in Canada"; "French is not currently supported." Source: Google Canada blog, week of October 28, 2024. Confidence: Verified.

A second Google surface, AI Mode — a dedicated AI-answer tab distinct from the inline AI Overview — launched in Canada on August 21, 2025, English only, announced by Robby Stein (VP of Product, Google Search). It appears as a tab, not as a forced replacement of the blue links.

Quote (Robby Stein, Google): "Today, we're beginning to roll out AI Mode in Google Search for users in Canada." Source: Google Canada blog / Robby Stein announcement, August 21, 2025. Confidence: Verified.

Paid surfaces caught up next. Ads inside AI Overviews reached Canada on December 19, 2025, among 12 total English-language countries, via a quiet update to Google's help documentation — not a formal announcement. Sensitive verticals (adult, alcohol, gambling, finance, healthcare, politics) are excluded.

Source: Google help-doc edit detected by the SEO industry, December 19, 2025; no formal Google press release. Confidence: Verified (industry-tracked from Google's own doc edit).

The four surfaces — AI Overviews (inline summary), AI Mode (dedicated tab), Perplexity (live citation-first engine), and ChatGPT search (retrieval-augmented model output) — together define the citation target space. Microsoft Copilot sits alongside them with a smaller share. The critical mechanical point is that they do not share an index: AI Overviews reuses Google's existing index via what is documented as query fan-out / FastSearch, while ChatGPT and AI Mode draw from a wider pool.

Claim: AI Overviews specifically correlate with traditional rankings more than other surfaces, because Google reuses its index via "query fan-out" / FastSearch. ChatGPT and AI Mode draw from a wider pool. So ranking #1 neither guarantees nor is required for citation. Source: Industry analysis (BrightEdge, Ahrefs, vendor commentary, 2025–2026). Confidence: Industry-consensus.

The operational consequence is precision in language: rank is necessary-but-not-sufficient for AI Overview citation, and is neither necessary nor sufficient for ChatGPT or AI Mode citation.

AI Overview prevalence — the published range

There is no single authoritative figure for how often an AI Overview appears. Independent and vendor measurements diverge by definition and sample, and any tactical brief that cites one number as "the" prevalence is overstating the evidence.

The independent anchor is Pew Research Center's July 22, 2025 field study, which analysed real browsing data from 900 US adults across 68,879 Google searches in March 2025.

Claim: "Some 18% of all the Google searches in our study generated an AI summary as part of the search results." Source: https://pewresearch.org/short-reads/2025/07/22/ — Jul 22, 2025. Confidence: Verified (independent research org; no commercial incentive).

The exposure-side companion figure: 58% of those US adults conducted at least one Google search in March 2025 that produced an AI-generated summary — AI summaries are routine, not edge-case, even when the per-query rate sits at 18%.

Source: Pew Research Center, https://pewresearch.org/short-reads/2025/07/22/ — Jul 22, 2025. Confidence: Verified.

Vendor trackers — SEO platforms measuring their own keyword universes — report higher numbers and a rising trend. Semrush, tracking 10M+ keywords, found AI Overview prevalence rose from 6.49% in January 2025 to a peak of ~24.61% in July 2025, settling at ~15.69% by November 2025.

Source: https://semrush.com/blog/semrush-ai-overviews-study — 2025/26. Confidence: Single-source (vendor — SEO platform with incentive to show large numbers). Caveat: Semrush's keyword sample is not the same population as Pew's real-user search log.

BrightEdge reported AI Overview share of tracked queries grew from approximately 30% to 48% between February 2025 and February 2026 — a 58% year-over-year increase.

Source: BrightEdge, "AI Overviews at the One-Year Mark" — brightedge.com (Feb 2026). Confidence: Verified. Caveat: "Tracked queries" is a vendor-selected sample; the 48% is not directly comparable to Pew's 18% of all real-user searches.

Conductor (21.9M queries) sat between the two at 25.11%, up from 13.14% in March 2025.

Source: conductor.com — 2026. Confidence: Single-source (vendor).

The 16–48% spread across these sources is largely definitional — different keyword samples, "tracked queries" versus all queries, and the inclusion of niche or long-tail terms. The direction (rising) is consistent across every source; the level needs an anchor-plus-range treatment.

Gap: AI Overview prevalence has no single authoritative number. Independent (Pew 18%, Mar 2025) and vendor (16–48%, 2025–26) figures diverge by definition and sample. Cite as a rising range, not a point estimate. Confidence: Verified (gap-in-evidence).

The encyclopedic citation pattern that follows from this is to anchor on Pew and frame the vendor figures as a range, as documented in Editorial discipline and sourcing.

The behavioural consequence — clicks and session-ends

The strongest single statistic for "AI Overviews halve clicks on the queries they appear on" is also from Pew's field study. When an AI summary appeared, users clicked a traditional search result link in 8% of all visits; without an AI summary, they clicked in 15% — nearly twice as often. Only 1% of visits to pages with an AI summary included a click on a link inside the summary itself.

Quote (verbatim): "Users who encountered an AI summary clicked on a traditional search result link in 8% of all visits. Those who did not encounter an AI summary clicked on a search result nearly twice as often (15% of visits)… This occurred in just 1% of all visits to pages with such a summary." Source: Pew Research Center, July 22, 2025. Google publicly disputed the methodology as "not representative." Confidence: Verified (highest-quality field-based dataset on this question).

The session-end behaviour is consistent: 26% of users ended the search session after seeing an AI summary, versus 16% without.

Source: Pew Research Center, https://pewresearch.org — Jul 22, 2025. Confidence: Verified.

Together these three numbers describe a market in which the search session increasingly resolves on the results page itself. The strategic implication — and the foundation of every subsequent tactical pattern — is that being the cited source matters more as raw click volume falls. Optimising for rank without optimising for citation eligibility leaves the dominant outcome unaddressed.

Rank-versus-citation decoupling — four independent vendor datasets

The defining tactical fact of 2026 GEO is that the URLs cited inside generative answers are largely not the URLs that rank at the top of the conventional results. Four independent vendor measurements describe the gap and the rate at which it is widening.

BrightEdge (February 12, 2026): "Only about 17% of sources cited in AIOs also rank in the organic top 10. That number has been remarkably flat… Roughly 5 out of 6 AIO citations are pulling from content that isn't on page 1."

Source: brightedge.com — Feb 12, 2026. Confidence: Industry-consensus.

Ahrefs (863k keywords / 4M URLs): page-1 overlap with AI Overview citations dropped from 76% in July 2025 to 38% in 2026 — a halving in six to seven months.

Source: ahrefs.com — 2026. Confidence: Industry-consensus (consistent direction with BrightEdge and Moz). Caveat: Ahrefs is a vendor; large-N keyword sample is internal.

Ahrefs (August 2025), measuring ChatGPT, Perplexity, Copilot and AI Mode together: approximately 80% of cited URLs do not rank in Google's top 100.

Source: ahrefs.com — Aug 2025. Confidence: Industry-consensus.

Moz (2026): 88% of Google AI Mode citations sit outside the organic top 10.

Source: moz.com — 2026. Confidence: Industry-consensus (corroborates BrightEdge, Ahrefs).

The pattern across the four datasets is unambiguous: the more "AI-native" the surface, the lower the overlap with the classic top 10. AI Overviews — which reuse Google's index via query fan-out — sit at the high end of overlap (17%); AI Mode, which is closer to a generative answer engine, sits at the low end (12% of citations are in the top 10, or 88% outside). ChatGPT and Perplexity, drawing from broader sources still, push the gap to 80% of citations falling outside the top 100 entirely.

The trajectory matters more than any single level. Ahrefs' 76% → 38% drop in six to seven months indicates the two systems are diverging rather than converging. Practitioner programs that report Google rank as the sole success metric have, by 2026, lost most of their explanatory power over the surfaces that resolve the customer's intent.

The operational rule that drops out of this evidence is to add AI citation and branded / direct traffic as parallel metrics alongside rank, rather than replacing rank.

Rule: Stop reporting Google organic rank as the sole success metric for client SEO and content programs. Add AI citation (BrightEdge / Ahrefs / Perplexity-style measurement) and branded / direct traffic as parallel metrics. Confidence: Industry-consensus across four independent vendors.

Freshness as a measurable signal

Freshness is the second tactical pattern with broad empirical support. The single best-evidenced datapoint is Ahrefs' large-N comparison of citation age versus organic top-10 age.

Claim: "The average age of URLs cited by AI assistants is 1,064 days, compared to 1,432 days for URLs in organic SERPs — 25.7% 'fresher'… ChatGPT is most likely to cite newer pages." (Ahrefs, Despina Gavoyannis, 16.975M citations across 7 AI platforms.) Source: https://ahrefs.com/blog/fresh-content/ — 2026. Confidence: Verified (primary measurement on a large sample). Caveat: Ahrefs sells SEO tools, but this is a direct measurement on its own crawl — vendor source, large-N primary data.

Two complementary vendor measurements bracket the operational cadence the data implies:

BrightEdge — pages updated within 60 days are ~1.9× more likely to appear in AI answers (2026). Source: brightedge.com — 2026. Confidence: Single-source (vendor).

AirOps — pages not updated in 90+ days are ~3× more likely to lose AI citations (2026). Source: airops.com — 2026. Confidence: Single-source (vendor). Caveat: Vendor self-reported; "lose citations" methodology not surfaced.

A third corroborates the direction:

Amsive — ~50% of AI-cited content is less than 13 weeks old (2026). Source: amsive.com — 2026. Confidence: Single-source (vendor).

Read together — Ahrefs' 25.7% freshness gap, BrightEdge's 60-day window, AirOps' 90-day cliff, Amsive's 13-week distribution — the published evidence places the operational refresh cadence for AI-citation-relevant pages at every 60–90 days. The qualifier "substantive" carries the load: Google's documented Query Deserves Freshness behaviour and the 2024 leaked Content API documentation both indicate that recency is a conditional ranking factor (strongest for time-sensitive queries) and that cosmetic date changes do not count.

Claim: Google's QDF behaviour and the 2024 leaked Content API documentation both indicate recency is a conditional ranking factor — strongest for time-sensitive queries, requiring substantive updates (not cosmetic date changes). Source: Google documentation history; 2024 Content Warehouse API leak coverage. Confidence: Industry-consensus.

A primary-source corroboration from Google's own documentation completes the picture: rich results are explicitly withheld from stale time-sensitive content.

Claim: Google's structured-data guidelines: "Provide up-to-date information. We won't show a rich result for time-sensitive content that is no longer relevant." Source: Google Search Central, "General Structured Data Guidelines" (accessed June 2026). Confidence: Verified — primary.

The encyclopedic rule that follows: pages where the working surface depends on currency, or where AI citation matters, are refreshed substantively at least every 60–90 days, and cosmetic date bumps do not count.

Body-text citations, quotations and statistics — the peer-reviewed lever

The single strongest piece of evidence in the entire GEO literature is a peer-reviewed paper from Aggarwal et al. (Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi) titled "GEO: Generative Engine Optimization," published at ACM SIGKDD KDD '24 (arXiv:2311.09735).

Claim (verbatim): "Including citations, quotations from relevant sources, and statistics can significantly boost source visibility, with an increase of over 40% across various queries." Top methods (Cite Sources, Quotation Addition, Statistics Addition) "achieved a relative improvement of 30–40% on the Position-Adjusted Word Count metric and 15–30% on the Subjective Impression metric," with "visibility improvements up to 37%" on the live engine Perplexity.ai. Source: Aggarwal et al., arXiv:2311.09735, published at KDD '24 — June 28, 2024 (v3). Confidence: Verified (peer-reviewed). Critical methodology caveat: These are edits to visible page text, NOT schema markup. Visibility was measured with the authors' own metrics on their GEO-bench (~10K queries, GPT-3.5 answer generator) plus a 200-sample Perplexity test — citation-share in synthesised answers, not real click traffic.

The paper's three operational findings carry separate weight:

The Position-Adjusted Word Count result — the harder of the two metrics, measuring actual word-share in the synthesised answer weighted by position — showed a 30–40% relative improvement from the Cite Sources, Quotation Addition and Statistics Addition methods.

Source: Aggarwal et al., arXiv:2311.09735 — Jun 28, 2024. Confidence: Verified.

The Subjective Impression result — a softer LLM-rated measure of how favourably the source comes across — showed a 15–30% relative improvement.

Source: Aggarwal et al., arXiv:2311.09735 — Jun 28, 2024. Confidence: Verified. Caveat: Uses an LLM as judge, so the metric carries the usual circularity caveats of LLM-as-judge methodology.

The live-engine validation — visibility improvements up to 37% on Perplexity.ai — is the only test against a shipping AI answer engine in the paper.

Source: Aggarwal et al., arXiv:2311.09735 — Jun 28, 2024. Confidence: Verified for the test result. Caveat: Sample of 200 on a single engine.

The numerous SEO vendor articles that repeat "Princeton GEO: 30–40% higher visibility from statistics / citations" all trace back to this one arXiv paper. They are repetition, not independent confirmation.

Source: Cross-vendor observation, 2024–2026. Confidence: Single-source (one primary, many echoes).

The methodology caveat is the lever for tactical interpretation. The peer-reviewed lift comes from edits to visible body text: adding named primary-source citations with dates, including direct quotations, and embedding hard statistics with attribution. It does not come from schema markup. The operational rule is therefore to invest editorial effort in body-text citations, quotations and statistics before investing more in schema, and to treat schema as eligibility-not-ranking — covered further in Technical SEO standards and structured-data discipline (2026).

Rule: When the goal is AI-answer citation, invest editorial effort in body-text citations, quotations and statistics before investing more in schema markup. The peer-reviewed evidence is on the body-text side. Confidence: Verified (peer-reviewed) for the body-text claim; no independent primary evidence exists for the schema-as-AI-visibility-lever claim.

Direct-answer-first page structure

A second body-text pattern with consistent practitioner support is opening top service and FAQ pages with a direct answer to the customer's question — pricing ranges for the area, named staff with credentials, real project specifics — inside the first 100–150 words, before any marketing preamble.

The dual rationale is that this serves the conventional E-E-A-T / Helpful Content regime (which rewards demonstrable first-hand experience) and the AI citation regime (which preferentially surfaces concrete, extractable answers and absorbs generic templated content without citing it). The behavioural cost of not being the cited source is the Pew 8%-vs-15% click pattern documented above.

Rule: On top service and FAQ pages, the first 100–150 words answer the customer's actual question directly — pricing ranges for the area, named staff with credentials, real project specifics. No marketing preamble. Confidence: Industry-consensus; grounded in Pew click data and the GEO paper's preference for concrete, extractable claims.

The applied pattern is to audit existing top pages, rewrite the first paragraph to lead with the answer plus specifics, and then measure AI Overview citation rate on the target queries — the measurement loop, not just the edit, is what makes the rule operational.

Where citability is strongest — and where it is not

The findability / citability argument is not uniformly strong across business types. The Whitespark Q2 2025 local-search study — measuring real Canadian / global local-SEO behaviour — is the single most useful query-mix datapoint.

Claim: AI Overviews appeared on 15% of simple local-intent queries (e.g., "tacos San Francisco") vs 92% informational (e.g., "how long does an eye exam take") vs 97% hybrid (e.g., "average cost of dental implants in Phoenix"). Local pack appeared on 93% of local-intent queries but only 6% of informational. Source: Whitespark Q2 2025 study. Confidence: Industry-consensus.

The implication is direct: buyers searching "[trade] near me" or "[city] roof repair" mostly still see the local pack and the Google Business Profile, not an AI summary. Informational queries are where AI Overviews bite.

Claim: The citability argument is strongest for businesses whose customers research questions (services, considered purchases, B2B), because AI Overviews and informational-query visibility concentrate there. It is weakest for purely transactional / local intent — AI Overviews appear on only a small share of e-commerce / shopping queries and local-pack-style searches. Source: Synthesis of Pew March 2025 data + industry analysis of AIO query mix (BrightEdge, Conductor 2025–2026). Confidence: Industry-consensus.

Trades, restaurants and local retail benefit most from the interactive and live-data capabilities documented in Faceted search and structured content for small-business sites — booking surfaces, availability, status. Service businesses and B2B benefit most from the structured and fresh capabilities — findability plus citation. The four-capability frame in the companion brief and the GEO patterns here are complementary, not competing.

The downstream tactical implication is that recommendations should be matched to the query mix the business actually faces, and the only way to know that mix is to measure it.

Monitoring as a tactical requirement

Because AI Overview prevalence is moving fast and is uneven across query types, recommendations grounded in last quarter's data go stale quickly. The published evidence — Semrush's 6.49 → 24.61 → 15.69% swing across 2025, BrightEdge's continued climb to 48%, Whitespark's 15-vs-92% local-vs-informational split — together makes the case that the operational program needs ongoing measurement on three dimensions for every priority keyword set: (a) whether an AI Overview appears, (b) whether the client is cited inside it, and (c) the local-pack position.

Rule: For every active client, track the priority keyword set on three dimensions: does an AI Overview appear, is the client cited inside it, and what is the local-pack position. Snapshot regularly, not just once. Confidence: Industry-consensus.

The applied pattern shifts effort based on what the measurement shows. If AI Overviews begin appearing on the client's core "near me" or commercial queries at scale (Whitespark's ~15% in Q2 2025), effort shifts toward being cited in AI answers and toward paid placement inside the AI Overview (available in Canada as of December 19, 2025, with sensitive verticals excluded). If informational blog traffic drops sharply — the Pew / behavioural pattern — that content is pivoted toward conversion-oriented, experience-rich pages.

Forward-looking — tools, brand mentions, and the limits of confidence

A weaker but worth-noting forward-looking claim is that AI answer engines may increasingly surface and cite useful tools, and that brand mentions and links correlate with AI-overview visibility. This is the most speculative claim in the published landscape — useful as forward-looking framing, not as a planning assumption.

Claim: AI answer engines (AI Overviews, AI Mode, Perplexity, ChatGPT search) may increasingly surface and cite useful tools, and brand mentions / links correlate with visibility in AI overviews — but this is early and contested. Confidence: Directional-Speculative. Caveat: Do not over-weight. Useful as a forward-looking line, not as a planning assumption.

If the pattern holds, it argues that linkable assets — calculators, structured data tools, queryable catalogues — strengthen rather than weaken in the AI-search era, because AI engines surface useful tools as citations rather than as ad slots. The mechanism is the same one that distinguishes structured records from prose:

Claim (synthesis): A structured catalogue is a body of records in which each item is described by defined, independent attributes. Because the attributes are exposed as data rather than buried in prose or pixels, the catalogue can be queried and filtered by visitors via faceted search, and read by machines — search engines, AI answer engines, and downstream tools. Source: Synthesis of USPTO structured-data patent definition, Integrate.io structured-vs-unstructured, Google structured-data mechanism examples. Confidence: Industry-consensus.

This is the same logic that anchors the broader working-surface thesis documented in Faceted search and structured content for small-business sites — structured, interactive, frequently-updated content is measurably more findable and more citable than static prose.

What the same evidence base says does not work

A handful of widely-repeated GEO tactics fail when measured against the published evidence. The encyclopedic record is worth keeping clean.

Single-vendor prevalence figures cited as "the level." Every vendor tracker has an SEO-platform incentive to show large numbers, and uses a different keyword sample. The 16–48% spread across Semrush, Conductor and BrightEdge is largely definitional. The published level is Pew's independent 18% (March 2025); the published range is the vendor data; citing a single vendor figure as the prevalence overstates the evidence.

Schema markup as the AI-visibility lever. No independent or primary evidence demonstrates that schema improves AI citation rates. The peer-reviewed lift documented in Aggarwal et al. (KDD '24) came from edits to visible body text. Schema covers eligibility for rich results — a separate downstream gate, not a citation lever — and is treated in Technical SEO standards and structured-data discipline (2026).

Cosmetic date bumps as a freshness signal. Google's leaked Content API documentation indicates substantive updates are required; QDF is a conditional ranking factor, strongest for time-sensitive queries. Changing only the visible "updated" date without underlying content change does not move the signal.

Google rank as the sole success metric. Across four independent vendor measurements, the URLs cited inside AI answers are largely not the URLs at the top of the conventional results, and the trajectory is widening rather than narrowing. Programs reporting on rank alone leave the dominant outcome surface unaddressed.

Repeating a vendor echo as confirmation. The "Princeton GEO: 30–40% higher visibility" claim widely repeated across the SEO press all traces back to the same Aggarwal et al. paper. Repetition is not independent confirmation; the underlying paper's methodology caveats apply every time the claim is restated.

Summary of the operational pattern set

The tactical patterns that survive scrutiny in the published 2026 evidence form a small, internally-consistent set: anchor AI-prevalence claims on Pew's independent 18% and frame vendor trackers as a 16–48% range; report AI citation and branded / direct traffic alongside Google rank rather than substituting for it; refresh AI-citation-relevant pages substantively every 60–90 days; lead the top 100–150 words of service and FAQ pages with the customer's answer plus specifics; invest editorial effort in body-text citations, quotations and statistics before adding more schema; match the chosen lever to the business's query mix (informational and B2B research lean on citation; local-intent leans on local pack and GBP); and measure AI Overview presence, citation, and local-pack position on the priority keyword set on a continuing basis.

Each of these patterns is grounded in at least one piece of independent or peer-reviewed evidence, corroborated by multiple vendor measurements pointing in the same direction, and stated with the methodology caveat the underlying study supports. The discipline of how those citations and caveats appear in published material — Source, Confidence, Caveat — is documented separately in Editorial discipline and sourcing, and the underlying discovery and indexing layer those patterns sit on top of is documented in How Google crawls, discovers, and indexes pages. The technical-SEO surface area — schema, robots directives, sitemaps, the eligibility-not-ranking framing — is documented in Technical SEO standards and structured-data discipline (2026). The cross-page companion that holds the foundational citation-rate research and the published-evidence catalogue is AI Overview citation patterns (GEO/AEO).