Google AI Mode vs AI Overviews: Two Surfaces, Two Link Strategies

They agree on the answer and disagree on the source. Why a 13.7% citation overlap means you need two playbooks — and how a UK brand earns a place in both.

Same answer, different sources: the data that forces two strategies

Google now answers the same question in two places that quietly disagree about who deserves credit. AI Overviews are the synthesised summary boxes sitting above the blue links on roughly a quarter of searches. AI Mode is the separate conversational tab — made the global default search experience at Google I/O in May 2026, powered by Gemini — that replaces ten links with a chat-style answer and follow-ups.

The instinct is to treat them as one target. The data says otherwise. Ahrefs’ analysis of 730,000 response pairs found the two surfaces reach the same conclusion about 86% of the time but share only 13.7% of their citations — they agree on what to say and disagree on where they found it. A separate study of 1,500+ queries found more than three-quarters of unique domains appeared in only one of the two. Visibility in one does not buy visibility in the other, which is the entire reason this article splits the playbook in two. It sits inside our wider link building strategies framework.

One asymmetry is your planning shortcut: if your brand appears in an AI Overview, Ahrefs found roughly a 61% chance it also appears in AI Mode’s longer answer — but the reverse is far weaker. Depth travels upward; brevity does not travel down.

Why does the same company produce two such different citation pools? Because the two surfaces are solving different problems. An AI Overview is a confidence play: Google only fires it when a clear consensus already exists across authoritative sources, and it wants the safest, most agreed-upon summary it can assemble quickly. AI Mode is a reasoning play: it assumes the user wants a thorough, multi-angle answer and is willing to spend more retrieval to build it. A consensus engine and a reasoning engine reach for different evidence even when they arrive at the same headline conclusion. That single insight explains almost every tactical difference that follows, and it is why “we rank well, so we’ll be cited” is no longer a safe assumption on either surface.

The stakes are not small. AI Overviews now reach an audience an order of magnitude larger than the standalone chatbots that dominate the headlines, and AI Mode became the default search experience at I/O in May 2026, which means the conversational surface is no longer a niche opt-in for early adopters — it is where a growing share of ordinary searches now resolve. For a UK brand, being absent from both is not a missed-upside problem; it is a slow erosion of the discovery channel that has underpinned organic growth for two decades.

The Two-Surface Citation Model

Use this as the spine for every page decision. Each surface is a separate channel with its own retrieval and its own win condition, and the value of naming the model explicitly is that it stops teams from doing “AI optimisation” in the abstract and forces a choice about which lane a given page is fighting in.

The strategic order is settled by the data: build for AI Mode depth first, then layer AI Overviews structure on top. Content engineered for fan-out extraction almost always also qualifies for overview selection; thin content built only for a summary box rarely survives the fan-out.

This ordering also protects your budget. Teams that chase the overview first tend to produce short, summary-shaped pages that win a chip on a handful of queries and then plateau, because there is no depth underneath for AI Mode to draw on. Teams that build genuine depth first earn AI Mode presence across a topic’s sub-questions and pick up overview citations almost as a by-product, because a deep, well-structured cluster contains plenty of liftable consensus capsules. Depth is the asset; the overview-ready capsule is a feature you extract from it, not a separate thing you build.

How each surface actually retrieves

Understanding the retrieval mechanics is what lets you stop guessing. The two surfaces differ less in their generation step — both use Google’s models to write the answer — and more in how they gather evidence before writing. Get the gathering step right and the citation tends to follow.

AI Overviews: consensus synthesis

An overview is triggered on high-confidence informational queries where a clear consensus exists across authoritative sources. Google fans the query into sub-queries, ranks sources for each, and synthesises a short summary with citation chips. Crucially, ranking still matters but no longer decides extraction: top-10 rankers supplied about 76% of overview citations in mid-2025 but only roughly 38% by early 2026. Ranking earns candidacy; passage structure and consensus earn the citation. The same answer-first discipline that wins classic position-zero applies here — see our guide to link building for featured snippets.

The behavioural shift behind overviews is what makes the citation valuable. Multiple independent studies agree on the direction even where they disagree on the magnitude: clicks fall sharply when an overview appears, with one large dataset showing organic click-through roughly halving on affected queries and a randomised field experiment measuring a 38% decline in outbound clicks. But the same research surfaces the opportunity hiding inside the scare: brands cited inside an overview earn materially more clicks than uncited brands on the identical query — one analysis put it at around 35% more organic and 91% more paid clicks. The citation has become the new ranking. You are no longer fighting for the blue link; you are fighting to be the named source inside the box that replaced it.

What earns the overview citation, in practice, is being the most liftable expression of the consensus. That means a crisp, accurate, answer-first capsule near the top of a page that already ranks, supported by schema that disambiguates your content, and free of the dense brand-narrative preamble that buries the answer the engine wants to lift. If your page makes Google work to find the two sentences that answer the query, a competitor whose page hands them over cleanly will win the chip.

AI Mode: query fan-out

AI Mode is a different machine. Gemini decomposes a single query into as many as 16 parallel sub-queries, each retrieving passages from separate sources at once, then stitches a long, reasoned answer with inline citations and memory across follow-ups. A question like “how does GEO work for B2B SaaS?” might fan out into the definition of GEO, how it differs from SEO, schema requirements, example cited content and FAQ structure — each a separate retrieval. That is why AI Mode cites a wider, deeper pool and names far more entities: it is assembling an answer from many passages, not summarising one consensus.

Two consequences follow. First, encyclopedic and reference content over-performs in AI Mode — Wikipedia appears in about 28.9% of AI Mode citations versus 18.1% in overviews — so a clear, consistent entity presence across reference sources pays off. Second, content must be chunked into independent, self-contained passages, because the engine extracts blocks, not pages. Dense, unbroken prose is the single most common reason a strongly-ranked page wins no AI citations.

The fan-out also changes how you should think about keyword research. A single commercial query is no longer one target; it is the hub of a dozen latent sub-questions the engine will go looking for. The brands that win AI Mode are the ones whose content already answers those adjacent sub-questions — not because they guessed the exact phrasing, but because they built genuine topical depth around the buyer’s real decision. This is why thin, single-page coverage of a topic loses badly here: when the engine fans out across eight sub-needs and your site answers two of them, it fills the other six with competitors, and your brand is outnumbered in the synthesised answer even if your one page was excellent. Depth is not a content-marketing virtue here; it is a retrieval requirement.

The entity finding deserves its own emphasis. AI Mode names roughly 2.5x more brands and people than an overview, which means it is constantly making decisions about which named entities belong in an answer. If your brand, your products and your authors are described inconsistently across the web — different names, different categorisations, contradictory facts — the engine has a harder time confidently placing you in that expanded entity set. Entity discipline is therefore not a nice-to-have for AI Mode; it is a precondition for being named at all.

It helps to picture the fan-out concretely. Imagine a UK buyer asking AI Mode, “which link building agency should a B2B SaaS company in London use, and how do I evaluate one?” A single overview might summarise a consensus listicle. AI Mode, by contrast, is likely to fan that into separate retrievals for what link building is, how agencies are priced, what good outreach looks like, red flags in vendor selection, UK-specific considerations, and comparison criteria — then weave a long answer that may cite a different source for each strand. A brand that has authoritative, liftable content on five of those six strands gets named repeatedly and shapes the recommendation; a brand with one strong sales page gets mentioned once, if at all, and watches competitors fill the rest of the answer. That is the entire commercial case for topical depth in a single illustration.

The shared foundation, then the splits

The good news for a stretched team is that the two surfaces share a foundation, so the bulk of the work compounds across both. The bad news is that the foundation is necessary but not sufficient — once you have it, the splits decide who wins. Do these shared fundamentals once and they pay on both surfaces:

• Answer-first capsules. Open the page and each section with a direct, quotable 2–3 sentence answer, then expand.
• Independent passages. Structure content as ~200–400 word self-contained sections under question-mirroring headings that extract whole.
• Schema and clean HTML. Article and FAQPage schema, server-rendered content, ungated authority pages.
• E-E-A-T and original value. Proprietary data, named authors, real experience — if an AI can fully replace your page in two sentences, it was not distinctive enough to cite.

Then split your effort:

• For AI Overviews: target the high-confidence informational and “best of” queries you already rank for; tighten the consensus answer and the schema.
• For AI Mode: build topic clusters around latent intents and pain points across the buyer journey, add entity breadth, and pursue encyclopedic/reference presence so you survive the fan-out.

A useful test when you are unsure which lane a page belongs in: ask what the user is trying to do. A user typing a settled informational query — “how to clean a suede jacket,” “best budget laptops 2026” — is in overview territory, and your job is to be the cleanest consensus answer. A user exploring, comparing or reasoning through a decision — the long, conversational phrasing AI Mode is built for — is in fan-out territory, and your job is to have depth across the decision. Most commercial topics have both kinds of query, which is why most priority topics deserve both a tight, schema-backed answer page and a deep cluster behind it. Mapping each priority query to a lane before you write is the single habit that keeps an AI-search content plan from drifting into generic “make it AI-friendly” busywork.

Five patterns that fail on both surfaces

Across content audits in early 2026, the same failure modes recur in pages that rank strongly yet earn no AI citations. They are worth checking against your own templates before you build anything new.

• Dense, unbroken prose. Long paragraphs with no headings or breaks cannot be extracted cleanly, so neither surface lifts them.
• The answer buried below the fold. Brand narrative or throat-clearing before the actual answer pushes the liftable sentence out of easy reach.
• Commodity content. If a two-sentence summary fully replaces your page, it was never distinctive enough to deserve a citation — originality is the moat.
• Thin topical coverage. One shallow page per topic gets outnumbered when the fan-out wants depth across sub-questions.
• Inconsistent entities. Contradictory names, categories or facts across the web make the engine hesitate to place your brand in the answer.

Off-page in 2026: information consistency, not just links

The biggest off-page shift is that Google’s AI weights claims it can verify across independent sources. Research into how large models ground answers points the same way: a claim that appears consistently across several credible, independent sources is trusted more heavily than the same claim made loudly in one place. The job has therefore moved from “acquire do-follow links” to “keep the same accurate claim about your brand live everywhere the model grounds” — your site, review platforms, industry publications, comparison content and community threads. Links still matter enormously as an authority signal: fundamental measures like Domain Rating, keyword coverage and backlink profile remain the strongest predictors of AI citation in study after study. But a link without consistent, corroborated claims around it is a weaker citation candidate than it used to be, and a strong, consistent claim with no authoritative links behind it still struggles to be trusted. The two reinforce each other.

Practically, that means your backlink profile and your brand-claim consistency now work as one system. Earn editorial links from authoritative, indexed sources, and make sure the facts those sources state about you match your own pages and your reviews. Diversify where the claim lives — community and reference ecosystems quietly feed the models even when they rarely appear as visible citations. For the data behind link-driven citation behaviour, see our link building statistics for 2026.

There is a measurable gap between what is visible and what is influential here. Analyses of large citation samples have found that community platforms can appear in only a fraction of a percent of visible citations while occupying a far larger share of an engine’s internal retrieval slots during query processing. The practical reading: a links-only view of off-page misses most of the retrieval surface. Corroboration on the sources the models actually read — even when those sources never show up as a cited chip — is now part of the job, and it is a job that classic link building is well-placed to do, because the same outreach relationships that earn a link can plant a consistent, accurate claim. The implication for a UK team with finite outreach capacity is to weight some of that capacity toward the highest-grounding sources in your niche — the publications, communities and reference pages the models lean on — rather than spreading every placement evenly by domain authority alone.

Measuring two surfaces without fooling yourself

The old scoreboard — rankings and sessions — no longer captures visibility, because a brand can shape an answer a user acts on without ever receiving a click. Expand the KPI set and, critically, report the two surfaces separately, since success in one does not imply success in the other.

• Citation share. Of the buyer queries you monitor, the percentage where your brand appears in the AI answer — tracked separately for overviews and AI Mode.
• Share of voice versus competitors. Your citation count against the combined competitor count across a fixed query set, so you can see whether you are gaining or losing ground.
• Entity mentions. How often your brand is named in AI Mode answers, cited or not — a leading indicator of how the model understands you.
• Revenue and assisted conversions. Because zero-click answers and background agents break last-click attribution, anchor reporting to outcomes, not raw traffic.

A realistic timeline helps set expectations: with consistently structured, depth-first content, early citations tend to appear within a week or two of publishing, while a meaningful share-of-voice improvement across priority buyer queries typically takes around three months of steady output. Treat the first month as instrumentation and baselining, not as the period in which you judge success.

A practical warning on tooling: because the two surfaces cite different sources, a single “AI visibility” number that blends them will hide exactly the signal you need. Insist on surface-level breakouts, and pair any third-party tracker with manual spot-checks — run your priority queries yourself, in both the overview and the AI Mode tab, and record who is named. The manual check is slow but it is the only way to see the answer your buyer actually sees, including the entity mentions that never resolve to a clickable citation and therefore never appear in automated citation counts.

What Google I/O 2026 changed — and what is coming

The May 2026 I/O event reset the baseline, so a 2026 strategy has to account for it rather than the 2024 launch picture. Three shifts matter most for link builders.

First, AI Mode went from opt-in experiment to the global default search experience, powered by the current Gemini generation, with Google rebuilding the search box for the first time in 25 years. That promotes the conversational surface — and its fan-out retrieval — from a fringe channel to a mainstream one, which raises the priority of the depth-first work this article argues for. Second, Google began placing ads directly inside AI answers, signalling that the AI surface is now a commercial property Google intends to monetise, not a neutral summary layer. For organic strategy that means the cited slot is about to get more contested, and earning it on merit becomes more valuable as paid placements crowd in around it. Third, autonomous and background agents entered the picture: searches that run in the background and agentic checkout flows that complete tasks without a visible session. These break the standard measurement model entirely — there is often no UTM, no click path and no session to attribute — which is precisely why the measurement section above leans on citation share and revenue rather than traffic.

The reassuring counterpoint, repeated across Google’s own commentary and independent analysis, is that search demand is at an all-time high. Clicks are being redistributed, not destroyed: people search more, and in longer, more conversational phrasing, which is exactly the shape AI Mode is built for. The opportunity inside the scary headlines is that the brands adapting their content to be cited — while competitors hesitate and keep chasing the vanishing blue-link click — capture the redistributed visibility. Starting now, while the surface is still being learned, is the largest advantage available in this transition, and it compounds: every quarter a competitor spends in denial is a quarter your entity and corroboration signals strengthen unopposed.

The UK angle

The two-surface dynamic lands slightly differently in the UK market, and four points deserve a British operator’s attention specifically.

• Click erosion hits UK publishers now. Independent research found clicks roughly halving when an overview appears (about 15% to 8%), and AI Mode pushing toward a ~93% zero-click rate. UK informational publishers — news, how-to, comparison and review sites that have long relied on top-of-page organic clicks — feel this first and hardest. The defensive response is not to fight the overview but to be the cited brand inside it, since cited sources still earn materially more clicks than uncited ones on the same query, and a citation also builds brand recall even on the zero-click majority.
• Watch Google’s self-preferencing. A large share of AI citations route back into Google properties and a majority of AI Mode citations can point at Google’s own result pages, so the open web competes with Google itself for the slot. For UK local and commercial intent this makes a well-kept Google Business Profile and Maps presence a primary discovery channel rather than an afterthought — for any brand with a physical or regional component, that profile is now first-class infrastructure, not a directory listing.
• Entity clarity in British English. Because AI Mode leans heavily on entity recognition, keep your brand, product and author entities labelled consistently with UK spelling and UK-specific framing — “optimise,” “organisation,” sterling pricing, UK regulatory references — so the fan-out reliably matches your pages to UK-intent sub-queries rather than treating you as a US-context near-match. Small consistency signals compound into confident entity placement.
• Diversify off Google. With only ~13.7% overlap between the two Google surfaces, and even less overlap across ChatGPT and Perplexity, a single-surface UK strategy is a single point of failure. The same depth-first, consistency-driven content that wins AI Mode is the content that travels best across every other answer engine, so building it once and distributing it widely is the most efficient hedge a UK brand can make.

Composite case study: a UK brand wins both lanes

Anonymised composite from typical UK B2B engagements; figures illustrate the pattern, not one named account.

A UK fintech ranked top-three on its core commercial terms yet was cited in neither surface, and its leadership had spent two quarters confused about why strong rankings were not translating into AI visibility. The audit made the cause obvious within a day, and it split cleanly along the two-surface model. For AI Overviews, its pages opened with three paragraphs of positioning and mission before reaching anything resembling a direct answer, so when Google looked for a liftable consensus capsule it found a competitor’s tighter page instead. For AI Mode, the fintech had exactly one page per topic, each comprehensive in word count but monolithic in structure, and almost no presence beyond its own domain — no consistent third-party description, no encyclopedic reference, thin review corroboration — so when the fan-out reached across a buyer question’s sub-topics, it retrieved rivals for most of them.

Over one quarter the team rebuilt the top 15 pages answer-first with independent ~300-word passages, added Article and FAQPage schema, built a tight topic cluster around buyer pain points, corroborated its key product claims across three UK industry publications and its review profiles, and tidied its entity into a consistent canonical description. By quarter end it was earning overview citations on its high-intent informational queries and surfacing in AI Mode answers across the cluster’s sub-topics — with no increase in raw page count. Depth plus consistency, not volume, moved the needle.

Two details from the engagement are worth generalising. The cluster work paid off in AI Mode faster than expected because the fan-out immediately found multiple of the fintech’s pages answering adjacent sub-questions, so a single AI Mode answer began citing the brand two or three times where it had cited it zero before. And the consensus work paid off in overviews precisely on the queries where the brand already ranked but had been invisible — the rewrite did not chase new keywords, it simply made the existing ranking liftable. The lesson for a constrained UK team is that the two lanes can often be won by reshaping assets you already have rather than commissioning a new content programme: candidacy was never the problem, liftability and entity clarity were.

Where this breaks: honest limitations

No optimisation removes the structural realities of these surfaces. Set expectations accordingly.

• The surfaces are converging in look, not retrieval. The I/O 2026 merge blurs the interface, but the underlying citation pools stay distinct — keep measuring them separately rather than assuming one report covers both.
• Measurement is broken by design. Zero-click answers, background agents and agentic checkout produce outcomes with no session and no last click. Track citation share and revenue, not raw sessions.
• Overlap figures are snapshots. The 13.7% overlap and entity ratios come from specific 2025–26 studies and will drift as models update; re-baseline quarterly.
• You compete with Google itself. A meaningful share of citations points back into Google properties, a structural disadvantage no optimisation fully removes.

Three jobs that now run together: SEO, AI SEO, GEO

It helps to name the three distinct jobs a 2026 strategy has to do at once, because teams often do one well and assume it covers the others. Classic SEO still earns the ranking and the indexability that make you a candidate — without it, neither AI surface can find you. AI SEO is the structural layer: answer-first capsules, independent passages, schema and entity clarity that let a model understand and extract your content cleanly. GEO, generative engine optimisation, is the authority-and-corroboration layer: original data, brand signals and consistent claims across the open web that make a model choose to cite you over an equally well-structured competitor.

Against the two surfaces, the three jobs distribute predictably. AI Overviews lean hardest on classic SEO plus the AI-SEO structural layer, because the engine is summarising a consensus it has already ranked. AI Mode leans hardest on AI SEO plus GEO, because the fan-out rewards depth and entity breadth more than raw position. The trap is treating these as sequential phases — “we’ll do SEO first, then AI later.” In practice they are one programme: a page rebuilt answer-first improves classic rankings, AI-SEO extractability and GEO citability simultaneously, which is why depth-first, consistency-driven work is so efficient. You are not buying three deliverables; you are doing one piece of work well enough that all three engines reward it.

For a UK link builder, the reframing is liberating rather than threatening. The relationships, outreach skill and editorial judgement that earn quality links are exactly the capabilities that plant consistent claims, secure corroboration and build the entity signals these surfaces reward. The discipline has not been replaced; it has been pointed at a new outcome — being named in the answer — and the operators who already do honest, authority-led link building are best placed to win it.

Your Monday-morning action plan

1. Track the two surfaces separately. Add AI Overview and AI Mode citation share to your reporting as distinct metrics alongside your monitoring tools.
2. Audit your top 20 commercial pages for buried answers; rewrite each with an answer-first capsule and ~300-word independent passages.
3. Pick one priority topic and build it into a cluster around latent buyer intents to feed AI Mode’s fan-out.
4. List your three most important product claims and make them identical across your site, reviews and at least two UK industry sources.
5. Strengthen one entity reference (consistent canonical description; pursue or correct an encyclopedic mention) to lift AI Mode presence.
6. Spot-check both surfaces monthly on your top 20 queries and log who gets cited where, then close the largest gap first.