Skip to content

GEO/AEO Benchmarks 2026: The Data on AI Search Impact

GEO/AEO Benchmarks 2026

TL;DR: AI Overviews now appear in 48%+ of queries and reduce organic CTR by 58-61%. But cited brands see 35% higher CTR. The instability is extreme — 40-60% of cited sources change month-to-month. ChatGPT referral traffic converts at 15.9% vs Google’s 1.76%. May 2026 update: Google AI Mode hit 75M daily users with 92-94% zero-click rate; AI Overviews now appear in 89% of brand searches. The load-bearing 2026 finding: 96% of AI Overview citations come from sources with strong E-E-A-T signals, and brand mentions correlate 3× more strongly with AI Overview visibility than backlinks (0.664 vs. 0.218). Domain Authority predicts less than 4% of AI citations.

The Headline Numbers

MetricValueSource
US population using AI search31.3%eMarketer
Google searches showing AI Overviews48%+BrightEdge
Organic CTR drop when AI Overviews present58-61%Ahrefs, Digital Bloom
AI Overview queries ending without any click93%Position Digital
Monthly churn in AI citation sources40-60%eMarketer
ChatGPT conversion rate vs Google organic15.9% vs 1.76%Position Digital

AI Overview Prevalence

AI Overviews have expanded dramatically:

MetricEarly 2025April 2026Change
Queries showing AI Overviews6.49%13.14%++102%
BrightEdge tracked queries~20%48%++58% YoY
Some categories32.76%

Industry-specific AI Overview growth:

  • Real estate: +258%
  • Restaurants: +273%
  • Retail: +206%

★ Insight ───────────────────────────────────── What this means: If you’re in real estate, restaurants, or retail, AI Overviews are now the dominant search experience for your queries. Optimizing for them isn’t optional — it’s survival. ─────────────────────────────────────────────────


Click-Through Rate Impact

The Bad News

When AI Overviews appear, organic clicks crater:

ScenarioOrganic CTRChange
No AI Overview1.62%baseline
With AI Overview0.61%-61%
Position 1 with AI Overview-58%Ahrefs data

Even non-AIO queries weakened: CTR dropped from 2.74% to 1.62% (-41%) — the entire search ecosystem is shifting.

Primary anchor (added 2026-06-07). The CTR-collapse figures above are vendor data (Ahrefs, Digital Bloom). A named research institution now corroborates the direction and rough magnitude: the Pew Research Center clickstream study (July 2025) — real browsing behavior across 68,879 searches from 900 US adults — found users clicked a traditional result on 8% of pages with an AI summary vs 15% without (~47% reduction), and clicked the AI summary’s own citations on just 1% of visits. Pew’s measured ~47% is the institutional analogue to the vendor “58–61%”; treat the vendor decimals as directionally right but not independently verified. (Tier 1 institutional primary; verified CONFIRMED 3-0. Google disputes the methodology — a self-interested rebuttal. See seo/zero-click-strategy § calibration for the full primary-vs-vendor breakdown.)

The Good News

Being cited in AI Overviews changes everything:

Citation StatusOrganic CTRPaid CTR
Not citedbaselinebaseline
Cited in AI Overview+35%+91%

The paradox: Fewer total clicks, but citation = massive advantage.


The Referral Traffic Quality Paradox

Here’s the counterintuitive finding that changes the calculus:

Traffic SourceShare of TrafficConversion Rate
Google Organic~99%1.76%
ChatGPT Referral~1%15.9%

ChatGPT referral visitors convert at 9x the rate of Google organic visitors.

The Washington Post found AI platform visitors converted to subscriptions at 4-5x the rate of traditional search visitors.

★ Insight ───────────────────────────────────── What this means: The “organic traffic crisis” headlines miss the point. Yes, volume is down. But AI-referred visitors are dramatically higher intent. A brand getting 1,000 ChatGPT referrals may outperform one getting 10,000 Google organic visits. ─────────────────────────────────────────────────

Traffic Growth Rates

MetricRate
LLM traffic YoY growth+527%
LLM traffic growth vs organic165x faster
ChatGPT share of AI referral traffic87.4%
AI traffic as % of average website traffic1.08%

Small slice. Growing fast. Converts better.


Citation Instability: The 40-60% Churn Problem

This is the most underreported finding:

40-60% of cited sources change month-to-month across Google AI Mode and ChatGPT.

What this means in practice:

  • You can be cited today, gone tomorrow
  • Visibility is far less stable than organic rankings
  • Continuous optimization is required — not a one-time fix
  • AI recommendations show <1% chance identical brand lists appear across repeated queries

Why it happens:

  • AI models update frequently
  • Training data refreshes
  • Retrieval systems change
  • Competition for citations is dynamic

Where AI Gets Its Citations

Top Domains Cited by LLMs

DomainNotes
Wikipedia7.8% of ChatGPT responses
RedditConsistently top-cited
YouTubeVideo content cited
ForbesAuthority publication
G2Review aggregator

The 6.5x third-party effect: Brands are 6.5x more likely to be cited through third-party sources than their own domains.

Content Positioning

Where in content do citations come from?

Content SectionShare of Citations
First 30% of content (intro)44.2%
Remaining 70%55.8%

Implication: The glossary/geo-anchor pattern is data-backed. Your first paragraph does almost half the work.

Content Types That Get Cited

Content TypeShare of Citations
Listicles21.9%
Articles16.7%
Product pages13.7%
“Best of” lists43.8% (ChatGPT specifically)

What converts best: Case studies and pricing pages drive highest AI referral traffic. Top-funnel content declined significantly.


The Organic Traffic Crisis: Winners and Losers

Overall Market

MetricValue
US organic search traffic YoY-2.5%
Zero-click searches60%
Mobile zero-click77%
Median publisher traffic YoY-10%

Organic Click Share Compression by Category

CategoryBeforeAfterChange
Headphones73%50%-23 pts
Jeans73%56%-17 pts
Greeting cards88%75%-13 pts
Online games95%84%-11 pts

Text ads gained 7-13 percentage points simultaneously.

Site-Level Winners and Losers

SiteTraffic ChangeWhy
HubSpot-70% to -80%Broad TOFU content model failure
People.com+27%Brand destination effect
Men’s Journal+415%Content resilience outlier
Top 10 sites+1.6%Relatively protected
Top 100-10,000 sitesSharp declinesConcentrated pain

★ Insight ───────────────────────────────────── The HubSpot lesson: Generic top-of-funnel “ultimate guide” content is exactly what AI Overviews cannibalize. If your strategy is “rank for informational keywords,” you’re in the blast zone. Brand-destination sites and niche authorities are more resilient. ─────────────────────────────────────────────────


User Behavior Changes

Decision-Making Patterns

BehaviorPercentage
Users adopt AI’s top recommendation74%
Users select 3rd-ranked option10%
Users override AI based on brand recognition26%
AI Mode users accept AI shortlist without verification88%
Traditional search users build independent lists56%

Implication: Position 1 in AI recommendations is disproportionately valuable. Brand recognition still matters — it’s the main override signal.

Query Behavior

MetricValue
ChatGPT prompts with no matching traditional keywords65-85%
Commercial intent prompts triggering web search53.5%
Informational intent prompts triggering web search18.7%
Time spent in AI Mode49 seconds
Time spent in AI Overviews21 seconds

Accuracy and Trust Issues

IssueRate
AI Overviews accuracy85-91%
AI Overviews containing false information9-15%
ChatGPT broken link rate2.38% (404 errors)
ChatGPT broken links vs Google3x higher

Implication: AI citations aren’t always accurate. This creates an opportunity — authoritative, accurate sources that AI can trust will be favored over time.


Budget Allocation Framework

Based on eMarketer recommendations for GEO/AEO investment:

CategoryAllocationFocus
Core SEO40%Foundation remains important
Digital PR25%Third-party citations matter more
Data & Reporting20%Citation tracking, Share of Model
Training10%Team capability building
Experimentation5%Test new tactics

Key quote from the research: “The overlap with what we’ve been doing in the SEO space and digital marketing before AI search existed is very, very strong.”


May 2026 update — Google AI Mode adoption + E-E-A-T as binary filter

Three substantial data shifts since this page’s original April 27 publication, all of which reshape the strategic picture:

Google AI Mode adoption (75M daily users; 92-94% zero-click)

Google AI Mode shipped to general availability in May 2026 and reached 75 million daily users within weeks. AI Mode usage among Google search sessions grew roughly 4× in 2 months (from ~0.25% in early May to over 1% by early July). The structural difference vs. traditional Google search:

MetricTraditional GoogleGoogle AI Mode
Zero-click rate35–46% (depending on AI Overview presence)92–94%
Average query length4.0 words7.22 words (almost 2× longer)
Average session lengthshorter49 sec (77 sec for brand-comparison queries)
External-domain visits per session~30–60%6–8%
Sources displayed per response10+ organic results1–3 sources cited

The AI Mode shift compresses the source-selection bottleneck dramatically. Where traditional Google offered 10 ranked organic results, AI Mode cites 1–3 sources per response. Citation share is now the metric that matters in this surface — see glossary/share-of-model for the competitive measurement framework.

E-E-A-T is now a binary AI visibility filter (not a soft ranking signal)

The biggest mechanism shift since April 2026. 96% of AI Overview citations come from sources with strong E-E-A-T signals. AI search engines use E-E-A-T as a binary gatekeeping filter — pages without strong E-E-A-T signals are not eligible for citation regardless of content quality.

The specific signals that load-bear E-E-A-T in 2026:

  • Earned media third-party validations (Forbes, industry publications, established outlets) — 90% of AI citations come from these sources, with citation value lasting 18–24 months after publication
  • Author-entity verification — consistent publishing depth, real author identity, proof points, and citation history. “Author-entity verification is now the load-bearing E-E-A-T mechanism in 2026”
  • Wikipedia presence and accuracy — disproportionately weighted in AI training data
  • Author + publisher schema — supports author-entity verification (a real mechanism); but schema markup itself has no measured effect on AI citations (see calibration below) — it’s not an AI-visibility lever
  • Topical authority depth (deep coverage of a specific topic) — outperforms broad coverage of many topics (content volume/length correlates with ≈nothing — 0.04)

The 2026 correlations finding inverts a 15-year SEO assumption. Ahrefs’ June 2026 study (n=75K brands) ranks the factors — YouTube mentions are the single strongest, above unlinked brand mentions, far above backlinks and Domain Rating:

SignalCorrelation with AI visibility
YouTube mentions0.737 (strongest)
Branded web mentions (unlinked)0.656–0.709
Domain Rating0.266–0.326
Backlinksweak
Site page count0.194 (≈ none)

Earned co-occurrence beats the link graph. AI engines weight aggregate brand-name + topic co-occurrence across the web — especially on YouTube — more heavily than link signals. A mention in a respected publication (or a YouTube video) outweighs dozens of self-published backlinks.

Calibration (2026-06-09). These are single-vendor, correlational figures (Ahrefs; “correlation isn’t causation”). The YouTube correlate may be confounded (big brands are on YouTube and get cited). The schema null is the one quasi-experimental result. Full grading: seo/ahrefs-ai-search-studies-2026.

Domain Authority predicts less than 4% of AI citations. The DA score the SEO industry has spent 15+ years gaming is no longer the load-bearing signal — earned media and brand mentions are.

The strategic implication: PR strategy is now SEO strategy. Earned media investment that 2020-era SEOs would have considered “brand work” now drives measurable AI-citation outcomes with 18–24 month compounding effects.

What this means for the dual mandate

The May 2026 data confirms what the April page set up: the 2026 operating model is dual-mandate — optimize for both traditional rankings AND AI engine citations. The two surfaces overlap (~38% of pages cited in AI Overviews also rank in top 10 in traditional Google, and that overlap is dropping) but require partially different strategies:

SurfacePrimary signals (2026)Primary metrics
Traditional GoogleBacklinks + on-page + technical SEO + content qualityPosition + organic clicks + impressions
AI engines (Overviews, AI Mode, ChatGPT, Claude, Gemini)E-E-A-T off-site validation + brand mentions + topical authority + structured citable claimsCitation share + share of model + branded search volume

The dual mandate is covered in operational detail in seo/zero-click-strategy.


Late-June 2026 update — measurement lands, and two trust/citation signals

Three developments since the May update. One is a genuine measurement primitive; two are signal-shifts that must carry credibility caveats.

Google Search Console gets a Generative-AI performance report (June 3, 2026)

Google launched dedicated Search Console reports for impressions inside generative-AI features — AI Overviews, AI Mode, and generative features in Discover. This is the first time site owners get a separate view of their presence in AI surfaces rather than having AI traffic blended into the main Search performance report.

The load-bearing caveat: at launch the report surfaces impressions-based metrics only (impressions, pages, countries, devices, dates) — clicks and CTR are excluded, with more metrics promised “over time.” Rollout is phased, starting with a UK subset. So it measures visibility (were you shown in an AI answer), not value (did anyone click). Still, it’s the first official, non-vendor instrument for tracking the citation-share metric this page has argued matters — pair it with glossary/share-of-model. (3-0; Google Search Central blog + Search Engine Journal + Google help docs + ppc.land.)

AI-search trust is declining year-over-year (Fractl survey, Q2 2026)

A repeated-questions YoY consumer survey (Fractl, Q2 2026, n=1,008 US consumers + 150 marketers) found AI-search “helpfulness” trust fell sharply: 82% found AI search more helpful than traditional results in 2025 vs only 54% in 2026 (a 28-point drop), and the skeptic camp grew ~sixfold (3% → 17%). For product recommendations specifically, consumers trust Google (39%) roughly 3× more than Reddit (15%) or AI tools (14%).

If it holds, this complicates the “everyone’s switching to AI search” narrative: adoption is rising (per the May data above) while trust in AI answers for purchase decisions is falling — the two aren’t moving together. It also tempers the seo/zero-click-strategy urgency slightly: for high-consideration purchases, the human still wants a trusted human/branded source to verify against.

Credibility tier: industry survey, vendor-interested, not peer-reviewed. Fractl is a digital-PR/AEO agency with a commercial interest in “brand visibility” consulting, and the relay (Search Engine Land) is a co-author secondary (presented at SMX Advanced 2026). Faithful descriptive restatement of a current survey — ingest as directional, not fact. (Verified 3-0 that the figures are reported as stated; the verification is of attribution, not of ground truth.) Update (2026-07-09): independent data now confirms the trust level is low (Reuters Institute, 20% global — see the early-July section below) but the decline trend remains uncorroborated outside Fractl. Update (2026-07-12): independent longitudinal data now exists for adjacent constructs — systems-trust declining (KPMG × Melbourne), product optimism rising (Ipsos/Stanford HAI); see the mid-July section. The AI-answer-trust decline specifically remains vendor-only.

A GEO-at-scale preprint — corporate sites dominate citations; sentiment is volatile

A single-author preprint (arXiv 2606.20065, June 18 2026; 100,000+ AI-search responses across 100+ brands) reports: ~78% of citations in AI answers go to corporate websites; among non-corporate sources, YouTube leads (ahead of Reddit, editorial media, Wikipedia — consistent with the YouTube-correlate finding above); and brand sentiment is far more volatile than brand presence — positive/negative framing flips ~6.7× more often than whether a brand is mentioned at all. The sentiment-volatility point is the genuinely new angle: monitoring whether you’re cited isn’t enough; how you’re framed swings round-to-round.

Credibility tier: single-author, non-peer-reviewed preprint by an author affiliated with Ranqo, a commercial GEO vendor — a conflict of interest, and the specific figures are single-sourced (secondary echoes trace back to the same paper). Ingest only as a labeled vendor preprint. Do-not-cite: the companion brand-tier-stratification claim from the same paper (73%/44%/11% visibility by brand size) was REFUTED (1-2) in verification — do not repeat it.

Early-July 2026 update — the independent trust data arrives (with a level-vs-decline distinction)

The late-June update above flagged an honest gap: the Fractl trust-decline survey was vendor-interested, with no independent corroboration. Two genuinely independent sources landed in mid-June and were verified in the July 9 sweep. They corroborate low trust — but not the decline trend. The distinction matters.

Reuters Institute Digital News Report 2026 — trust in AI answers is low (June 16, 2026)

The Oxford Reuters Institute’s DNR 2026 (~48 markets, independent academic body, no vendor stake) directly measured trust in AI-chatbot answers:

MetricValue
Trust in answers from AI chatbots (global)20%
Trust in news overall (comparison baseline)37%
Trust in AI-chatbot answers, UK (lowest market)6%
Trust among people who actually use chatbots44%

The load-bearing caveat: this is a LEVEL measurement, not a trend. It confirms trust in AI answers is low — roughly half the (already low) trust in news — but it does not corroborate Fractl’s 82%→54% decline, which remains vendor-survey-only with no independent longitudinal confirmation. Note also the user/non-user gap: among actual chatbot users trust more than doubles (44%), so part of the low global number is unfamiliarity, not verdict-after-use. (Tier 1 independent academic primary; verified 3-0, figures fetched verbatim from both DNR pages.)

Reuters Institute DNR 2026 — click-through, with two bases that must not be conflated

The same report measured self-reported click-through from AI answers — and the two denominators tell opposite-sounding stories:

  • 42% of people who use AI chatbots FOR NEWS say they always/often click through to original sources — between social media (36%) and search (44%). Chatbot users aren’t unusually click-averse.
  • Only 4% of ALL respondents always/often click through from chatbots (vs 19% from search, 17% from social) — because chatbot-news users are only ~9–10% of the population. The two figures are arithmetically consistent; quoting either without its base misleads.

Self-report caveat: these are stated propensities, not measured behavior — weak evidence next to behavioral clickstream studies (Pew’s measured 8% CTR). This page’s existing self-report-vs-behavior warning (seo/zero-click-strategy § calibration) applies in full. Useful mainly as a nuanced counterweight to the most extreme “nobody ever clicks out of AI” claims. (Verified 3-0; the report itself flags recall/social-desirability bias.)

Pew “Americans and AI 2026” — adoption and sentiment context (June 17, 2026)

Pew Research Center (N=5,119 US adults, fielded Feb 2026): 42% of US adults use AI chatbots for information searching — the top use case; four-in-ten expect AI to have a negative societal impact over 20 years; 63% say AI is advancing too quickly. This is adoption + general-sentiment data, not a trust-in-AI-search metric — the report contains no direct trust-in-results question, so it corroborates adjacent skepticism, not the Fractl decline. Fieldwork was February, so the 42% adoption figure is likely a floor. (Tier 1 independent pollster; verified 3-0.)

Where this leaves the trust question: adoption rising (Pew 42%), trust in answers low (Reuters Institute 20%), and the decline claim still resting solely on Fractl. Until independent longitudinal data appears, cite the trust story as “independently confirmed LOW, vendor-reported FALLING.” (Superseded by the mid-July section below — independent longitudinal data has now been located, and it complicates rather than settles the question.)

Mid-July 2026 update — independent longitudinal trust data lands (and the picture is construct-dependent)

The early-July section closed on an open target: no independent source tracked AI trust over time. The July 12 sweep located two — and they point in opposite directions, because they measure different things. The honest framing is now construct-dependent, not a single “trust is declining” narrative.

KPMG × University of Melbourne — trust in AI systems is declining (2022→2024)

The KPMG × University of Melbourne global study Trust, attitudes and use of AI (48,000+ respondents, 47 countries, fielded Nov 2024–Jan 2025; university-led design and analysis, not a vendor survey) tracks a 17-country panel against its 2022 wave:

Measure20222024Spread
Perceived trustworthiness of AI systems63%56%declined in 13 of 17 countries
Willingness to rely on AI52%43%declined in 12 of 17 (largest: Japan 43%→21%, Brazil 67%→53%)
Employee-reported organizational AI use34%71%adoption roughly doubled

The decline happened while adoption doubled — the authors’ own interpretation is “more exposure, more considered trust,” explicitly ruling out unfamiliarity-driven distrust. The trustworthiness construct includes the ability “to provide accurate and reliable output,” making it the closest non-vendor analogue to answer-trust currently available. (Tier 1 — university-led primary; all three figures verified 3-0 verbatim against the report PDF.)

Two caveats that must travel with any citation: (1) the construct is trust in AI systems broadly, not AI-search answers specifically; (2) the panel endpoint is late-2024 fieldwork — it corroborates the direction of the Fractl trend but cannot confirm 2025→2026 movement.

Stanford HAI AI Index 2026 — general AI-product optimism is rising (the counterweight)

The AI Index 2026 public-opinion chapter (relaying the Ipsos series, 2022–2025, 30 countries, n=23,216 in the 2025 wave) moves the other way: benefits-outweigh-drawbacks rose 55% (2024) → 59% (2025), while nervousness also rose (to 52%). The chapter’s embedded trust items are roughly flat: trust-companies-to-protect-data 54%→53%→53%; trust-AI-not-to-discriminate 50%→47%→48%. Fieldwork ended April 2025, so the series cannot rule out a later decline. (Tier 1 academic relay; verified 3-0 against the web chapter, 2-1 against the chapter PDF.)

Do-not-cite caveat: the framing of the Ipsos AI Monitor as a standalone “annual tracked trust instrument — exactly what the wiki lacks” was REFUTED (1-2) in verification. Cite the series only via the Stanford HAI AI Index 2026 chapter, and only for the optimism/nervousness trend plus the roughly-flat trust items, with the construct caveat (general product sentiment, not answer-trust).

Source audit — Edelman 2026 checked and eliminated

The 2026 Edelman Trust Barometer Global Report contains no longitudinal trust-in-AI measure: its single AI-specific slide reuses point-in-time data from the separate 2025 Flash Poll, and the widely floated “32% US / 72% China / 49% global trust AI” figures belong to the 2025 barometer, not the 2026 report. Recorded here to prevent re-investigation and mis-citation — do not cite the 2026 Edelman report as an AI-trust trend source. (Verified 3-0 against the full 49-page PDF.)

The construct-dependent bottom line

ConstructSource (independence)WindowDirection
Trust/reliance in AI systemsKPMG × Melbourne (independent, university-led)2022→2024Declining (63%→56%, 52%→43%)
General AI-product sentimentIpsos via Stanford HAI (independent academic relay)2022→2025Optimism rising (55%→59%); trust items flat
Trust in AI-search answers — levelReuters Institute DNR (independent academic)2026 point-in-timeLow (20% global)
Trust in AI-search answers — trendFractl (vendor-interested)2025→2026Falling — still vendor-only

Cite the trust story as construct-dependent, never as a flat “trust is declining”: systems-trust declining (independent, through 2024), product optimism rising (independent, through early 2025), answer-trust low (independent, 2026 level) — and the answer-trust decline specifically still has no independent longitudinal series reaching into 2026. That narrower gap stays open; the next test is whether KPMG/Melbourne or Ipsos fields a wave covering 2025–2026.


Tactical Recommendations (Data-Backed)

1. Optimize the First 30%

44.2% of citations come from intros. Use the glossary/geo-anchor pattern: answer the primary question in sentence one.

2. Create Citable Formats

  • Listicles: 21.9% of citations
  • “Best of” lists: 43.8% of ChatGPT citations
  • Tables and structured data

3. Build Third-Party Presence

6.5x citation advantage from third-party sources. Priority platforms:

  • Reddit (high citation rate)
  • Review sites (G2, TrustPilot)
  • Industry publications

4. Focus on Bottom-Funnel Content

Case studies and pricing pages drive highest AI referral conversion. Top-funnel informational content is getting cannibalized.

5. Monitor Citation Stability

40-60% monthly churn means continuous optimization, not one-time fixes. Track your “Share of Model” regularly.

6. Test Across Multiple AI Systems

Different models have different biases. What works for ChatGPT may not work for Gemini or Claude.


Key Takeaways

  • AI Overviews in 48%+ of queries — this is now mainstream
  • 61% CTR drop when AI Overviews present — but cited brands see +35%
  • 40-60% monthly citation churn — visibility is unstable
  • 15.9% conversion rate from ChatGPT vs 1.76% from Google — quality > quantity
  • First 30% of content = 44% of citations — intros matter most
  • Third-party sources cited 6.5x more — build presence beyond your site
  • HubSpot-style TOFU content failing — niche authority wins


Sources

Primary / institutional (verified 3-0)

SEO-vendor analyses