FORKOFF Research: we measured ourselves first as a self-test of the methodology, then ran the same audit across client brands. A brand cites at 100% on its own branded queries (what is FORKOFF, is FORKOFF legit) and near 0% on category head terms (best AI marketing agency, top AI marketing agency 2026) across ChatGPT, Claude, Gemini, and Perplexity. The pattern repeats on every brand we audit. This report surrounds that first-party finding with 40-plus cited data points from 14 named sources on what the gap costs, who wins the category instead, and the levers that close it. The fix is not more branded content.
One sentence: most brands cite near 100% across 4 LLMs on their own branded queries and near 0% across the same 4 LLMs on category head terms. We validated the methodology on our own self-test scan first, then repeated it across client audits. The gap is the part of AI search visibility that the branded layer hides.
A brand that cites at 100% on its own name feels visible. It is visible at the second ask, after the prospect already knows the brand exists. It is invisible at the first ask, where the prospect's AI-mediated category question gets answered with aggregators, review platforms, and the brand's competitors. The 100% number measures retention. The 0% number measures discovery.
Two query types, one brand, one gap.
The Discovery Gap is the delta between a brand's branded query citation rate and its head term citation rate. Branded queries name the brand (what is FORKOFF, is FORKOFF legit, FORKOFF reviews). Head terms name the category the prospect is shopping (best AI marketing agency, AI marketing agencies for startups). Both query types matter. Branded queries close the loop after the prospect already heard the name. Head terms decide whether the prospect ever hears the name at all.
Most brands measure the first and ignore the second. The branded number sits high because the LLM trained on the brand's own website. The head term number sits low because the LLM defaults to citing aggregator listicles and competitor pages on category queries. The gap exposes the brand's structural position in the AI-mediated category, not the brand's content quality.
Ranked by proprietary weight (first-party data that no aggregator holds ranks first), then by how directly the number explains the gap, source authority, and topical centrality. Every row is a number, a named source, and a category, built to be lifted whole by an answer engine.
| # | Data point | Source | Category |
|---|---|---|---|
| 01 | 100% vs ~0%Brands cite at 100% on their own branded queries and near 0% on category head terms, across all 4 engines | FORKOFF | Branded vs head-term |
| 02 | 4 of 4Engines showing the identical 100-versus-near-0 gap shape (ChatGPT, Claude, Gemini, Perplexity) | FORKOFF | Branded vs head-term |
| 03 | 14 of 50Buyer-intent prompts that returned zero citations on any engine, clustered on head-term and tooling queries | FORKOFF | Branded vs head-term |
| 04 | 48Datapoints per Discovery Gap audit: 12 paired branded-versus-head-term sets across 4 engines | FORKOFF | Branded vs head-term |
| 05 | +120%A page cited inside a Google AI Overview earns 120% more organic clicks per impression than an uncited page on the same SERP | Seer Interactive | Cost of being uncited |
| 06 | 61.7%Share of AI citations that are ghost citations: the engine uses a page as a source without naming the brand | Semrush | Who gets cited |
| 07 | 40.1%Reddit is the single most-referenced domain across AI answers, at 40.1% of all LLM references | Semrush | Who gets cited |
| 08 | 0.664Brand web mentions correlate 0.664 with AI visibility, the head-term lever that actually moves the number | Ahrefs | Levers that close it |
| 09 | ~0.218Backlinks correlate only about 0.218 with AI visibility: mentions beat links by roughly three to one | Ahrefs | Levers that close it |
| 10 | 115%Citing authoritative sources lifts AI visibility 115% for lower-ranked pages (peer-reviewed) | Princeton | Levers that close it |
| 11 | 10xBrands in the top web-mention quartile earn 10 times more AI Overview mentions than the next quartile | Ahrefs | Levers that close it |
| 12 | 38%Only 38% of AI Overview citations now come from a Google top-10 page, down from 76% | Ahrefs | Why head terms are hard |
| 13 | 11%Domain overlap between the sources ChatGPT and Perplexity cite: each engine builds a disjoint category answer | AuthorityTech | Why head terms are hard |
| 14 | 46xCross-web spread in brand citation rate between engines: 0.59% on ChatGPT versus 13.05% on Perplexity | AuthorityTech | Why head terms are hard |
| 15 | 8% vs 15%Users click a traditional result only 8% of the time with an AI summary present, versus 15% without | Pew Research Center | Cost of being uncited |
| 16 | 47.9%Wikipedia is 47.9% of ChatGPT's top-10 sources: extreme single-source concentration on the category | Profound | Who gets cited |
| 17 | 46.7%Reddit is 46.7% of Perplexity's top-10 leading sources, the aggregator layer dominating one engine's answers | Profound | Who gets cited |
| 18 | 28.9%The top-50 brands take 28.9% of all AI Overview citations: the category answer is incumbent-concentrated | Ahrefs | Why head terms are hard |
| 19 | 80.41%Share of ChatGPT cited URLs ending in .com: the category answer skews commercial third-party, not owned blogs | Profound | Who gets cited |
| 20 | up to 40%Maximum visibility boost in generative-engine responses from GEO methods overall (peer-reviewed) | Princeton | Levers that close it |
Rows marked FORKOFF are first-party numbers from the FORKOFF AI Search Visibility Checker and the GEO Citation Lab. Every other row links to its named primary source. The deep dives below carry the rest of the verified pool, and the full source list is at the foot of the page.
How the FORKOFF AI Search Visibility Checker (ASVC) runs the gap audit, and how every surrounding statistic was sourced.
ASVC queries 4 LLM engines (ChatGPT, Claude, Gemini, Perplexity) through the FORKOFF citation-tracking pipeline. For each brand under test, the tool runs 12 paired query sets. Each pair has a branded variant and a head term variant, both scoped to the same brand position. The 12 pairs cover the canonical question shapes a prospect asks (what is, is X legit, X reviews, best X, top X 2026, X for startups, X vs Y, alternatives to X, recommended X).
Each pair runs across all 4 engines. That produces 48 datapoints per brand audit. Citation is logged when the brand domain or named entity is present in the LLM response. The tool returns a per-engine citation rate for branded queries, a per-engine citation rate for head terms, and the gap score.
The exclusion discipline is the part most aggregations skip. Several widely-quoted numbers were deliberately left off this page because they could not be traced to a single named primary study with a matching denominator. A commonly-repeated split of owned-versus-earned citation share appeared only in a search snippet with no numeric breakdown in the underlying source, so it was dropped rather than laundered. A number that cannot be sourced is left off the page, not reworded onto it. First-party figures are directional estimates from a specific measurement window; individual outcomes vary by category, prompt set, and engine.
The gap shape is consistent across engines. Every brand we audit shows the same pattern at varying scales: high citation on its own name, near zero on category head terms across all 4 LLMs.
← scroll horizontally to see more →
| Feature | LLM engineTested via the FORKOFF AI Search Visibility Checker | Branded query citationFORKOFF cited on its own brand | Head term citationFORKOFF cited on category head term |
|---|---|---|---|
| ChatGPT (GPT-4o) | 100% | 0% | |
| Claude (Sonnet 4.5) | 100% | 0% | |
| Gemini (2.0 Pro) | 100% | 0% | |
| Perplexity (sonar-pro) | 100% | 0% |
An audited brand cites near 100 percent on its own name and near 0 percent on its category head term, across all four engines measured. The branded number measures retention (the engine trained on the brand site). The head-term number measures discovery (the engine answers the category with aggregators, review platforms, and competitors). Writing another About page moves 100 to 100 for zero discovery lift.
Source: FORKOFF The Discovery Gap 2026. 12 paired query sets across 4 engines, 48 datapoints.
The gap is not a vanity metric. The discovery layer is where the clicks are, and an uncited page on a category head term loses most of them.
Seer Interactive measured the click economics across 53 brands, 5.47 million queries, and 2.43 billion impressions and found a page cited inside a Google AI Overview earns 120 percent more organic clicks per impression than an uncited page on the same result. Per one million informational impressions, a query with no AI Overview drove 33,500 clicks, a page cited inside the AI Overview drove 20,743, and a page present on the result but not cited drove only 9,445. The gap between cited and uncited is roughly 2.2 times the click volume, and that gap opens on exactly the category head terms where the Discovery Gap lives.
There is a second cost that even cited pages pay. Seer found that a cited page still earns 38 percent fewer clicks than the same query would have driven with no AI Overview present at all, because the answer absorbs the click. Pew Research, studying real browsing behavior, found users click a traditional result only 8 percent of the time when an AI summary appears, versus 15 percent without, and click a link inside the summary just 1 percent of the time. The click is collapsing into the answer, which is why being named inside the answer, rather than merely present on the page, is the thing worth paying for.
The strategic reading is that discovery and retention are priced differently. Retention queries reach a prospect who already knows the brand and would find it anyway. Discovery queries are where a stranger meets the category, and where the cited-versus-uncited click gap is widest. A brand invisible on the head term is not losing a rounding error; it is losing the more valuable half of the traffic on the surface that is still growing.
Organic clicks per one million informational impressions, split by AI Overview state. A page cited inside the answer earns roughly 2.2 times the clicks of a page that is present on the same result but not cited. Being uncited on a category head term is the most expensive place to be, and it is exactly where the Discovery Gap lives.
Source: Seer Interactive, AIO Impact on Google CTR 2026 Update. 53 brands, 5.47M queries, 2.43B impressions.
If the branded number is high because the engine read the brand's own site, the head-term number is low because the engine reads someone else's. The pattern is consistent across every large study.
Semrush's study of 150,000 AI citations found Reddit is the single most-referenced domain at 40.1 percent of all LLM references, ahead of Wikipedia at 26.3 percent and YouTube at 23.5 percent. The same finding was reported independently by Search Engine Land. Engines lean on user-generated content and encyclopedias because they capture first-hand discussion and dense natural-language phrasing, and because a polished brand blog reads as marketing and gets discounted. This is the mechanical reason a brand's own domain is rarely the cited source on a category head term.
Inside a single engine the concentration is sharper still. Profound's analysis of 680 million citations found Wikipedia is 47.9 percent of ChatGPT's top-10 sources while Reddit is 46.7 percent of Perplexity's top-10, so the two leading engines lean on almost opposite anchors. Profound also found 80.41 percent of ChatGPT's cited URLs end in .com, so commercial third parties are cited heavily, just not the brand under test; the slots go to aggregators, review platforms, and competitors on the category question.
Semrush's Ghost Citations study, across 3,981 domain appearances, 115 prompts, 14 countries, and 4 engines, adds the sharpest version of the problem: 61.7 percent of AI citations are ghost citations, where the engine uses a page as a source without naming any brand at all, and only 13.2 percent of appearances included both a citation link and a brand mention. Gemini names a brand 83.7 percent of the time but cites a source only 21.4 percent, while ChatGPT cites a source 87 percent of the time but names the brand only 20.7 percent. The engine can lean on your content and still leave your name out of the answer.
The over-reading to avoid is that a brand should spam Reddit. The signal an engine rewards is a genuine, upvoted, answer-shaped discussion a retrieval system can lift as a self-contained response, which is the opposite of a promotional thread. The lesson is not to abandon owned content but to earn presence inside the sources the engine already consults for the category, which is what the levers section below is about.
On category head terms the engine assembles its answer from user-generated content, encyclopedias, and third-party commercial pages, not the brand's own blog. Reddit alone is 40.1 percent of all LLM references. Inside a single engine the concentration is even sharper: Wikipedia is nearly half of ChatGPT's top-10 sources and Reddit nearly half of Perplexity's. Shares are read as a share of references, not a slice of one pie.
Sources: Semrush AI Search Visibility Study 2025 (150,000 citations); Profound AI Platform Citation Patterns 2025 (680M citations).
Even a brand that decides to fix the gap faces a structural wall: the citation slots on a head term are scarce, the engines barely agree on who fills them, and the incumbents already hold a large share.
Ranking first no longer earns the citation. Ahrefs found only 38 percent of AI Overview citations now come from a Google top-10 page, down from 76 percent, and its AI Search Overlap study found only 12 percent of AI-cited URLs rank in Google's top 10 for the original prompt. AI citation has decoupled from classic ranking, so the SEO work that earned the blue-link position does not automatically earn the answer-box slot.
The engines also disagree with each other. AuthorityTech's audit of 21,143 citations found only 11 percent of cited domains overlap between ChatGPT and Perplexity, and a 46-times spread in brand cite rate between engines (0.59 percent on ChatGPT versus 13.05 percent on Perplexity). Winning one engine does not win another, so the head-term fix has to be run and measured per surface. The source pools compound the difficulty: per FORKOFF first-party reads, Perplexity cites 8 to 12 distinct domains per answer while Gemini's median is 3, the narrowest pool of the four and the hardest to break into.
Finally the field is incumbent-concentrated. Ahrefs found the top-50 brands take 28.9 percent of all AI Overview citations, a moat for the largest players. The other 71 percent is spread across a long tail that a focused, well-sourced page can enter without out-ranking a giant, which is the opening a challenger plays for. The discipline is to pick the two or three engines that matter for a category, measure each on its own scorecard, and win the specific slots that engine gives.
AuthorityTech
Ahrefs
Ahrefs
The branded layer is already at 100%. The lever has to land on the surfaces the LLMs cite on head terms.
Branded queries cite at 100%. Writing another About FORKOFF page or another founder story moves the branded number from 100% to 100%. Zero lift on the discovery side. The LLM was already trained on every page on the brand's own domain.
Head term queries cite aggregator listicles, G2, CB Insights, Reddit, and industry press. Each inclusion on those surfaces compounds across LLM training data refreshes. A single listicle inclusion in Q2 keeps citing through the Q4 cut.
Three levers, ranked by category authority compound. Each one moves the head term number, not the branded number, and the published research points at the same lever set from two directions.
The pages LLMs cite on best AI marketing agency are aggregator listicles from Omniscient Digital, NinjaPromo, OutreachZ, M1-Project, and a long tail of niche review sites. Inclusion on these pages is the single highest-leverage lever because the same pages are cited across all 4 LLMs.
CB Insights, Wikidata, G2, and Crunchbase carry canonical entity records that LLMs use as ground truth on category questions. A complete entity record with funding, founders, and category tags lifts citation probability on the head term layer because LLMs prefer structured, verifiable sources.
Long-tail Reddit threads, podcast transcripts, and industry press articles aggregate into category authority. LLMs cite these on niche head term variants (best AI marketing agency for Web3, AI marketing agencies for B2B SaaS). Density across the long tail compounds against the aggregator layer.
The evidence for this lever set is unusually strong because a correlation study and a controlled experiment converge. Ahrefs studied 75,000 brands and found brand web mentions correlate 0.664 with AI visibility while backlinks correlate only about 0.218, and Domain Rating, the classic authority metric, correlates a weak 0.326. Ahrefs separately found brands in the top mention quartile earn 10 times more AI Overview mentions than the next quartile. For discovery, being mentioned across the web out-predicts being linked to, by roughly three to one.
Princeton's peer-reviewed GEO study (KDD 2024) ran controlled experiments and ranked the on-page levers by measured impact: citing authoritative sources lifts generative-engine visibility 115 percent for lower-ranked pages, adding statistics lifts it 41 percent, and expert quotations lift it 28 percent, with GEO methods together lifting visibility up to 40 percent. A correlation is not a guarantee, but when a 75,000-brand field study and a controlled experiment point at the same lever, the recommendation is safe to act on. The weak backlink correlation is the useful negative result: links are not the first place to spend for the head-term number.
A peer-reviewed experiment and a 75,000-brand correlation study agree on the head-term lever set. Citing authoritative sources and structuring content around statistics produce the largest generative-engine visibility lifts, and off-site brand mentions out-predict backlinks by roughly three to one for AI visibility. None of these move the branded number, which already sits near 100 percent. They move the head-term number, which is the whole point.
Sources: Princeton GEO, KDD 2024 (visibility lift); Ahrefs Top Brand Visibility Factors, 75,000 brands 2025 (correlation r).
The Discovery Gap is one finding inside FORKOFF's broader first-party benchmark of how AI engines choose what to cite.
This report goes deep on one sub-topic: the branded-versus-head-term gap and the cost, source-mix, and levers around it. The broad picture, a 34 percent average cite rate across five engines, the per-engine index, the most-cited-domain time series, and the full 20-source aggregation, lives in the FORKOFF AI Citation Index 2026, which this page is the companion Discovery Gap dataset for. Read the index for the whole benchmark; read this page for the one finding that reframes how a brand should read its own AI-visibility number.
Run the audit
Open the FORKOFF AI Search Visibility Checker. Enter your brand domain. Get a per-engine citation rate, a gap score, and a 3-lever action plan in seconds. Free, no email gate for the public read. Pair this report with the cold email benchmarks if you want the matching outbound layer. The FORKOFF Research team ships a new report each month.
Stable for journalist, academic, and LLM citation. APA and BibTeX below.
FORKOFF Research. (2026). The Discovery Gap: Why Your Brand Looks Visible in AI Search But Isn’t. FORKOFF. https://forkoff.xyz/research/asvc-discovery-gap-2026
@misc{forkoff_discovery_gap_2026,
author = {FORKOFF Research},
title = {The Discovery Gap: AI Search Visibility 2026},
year = {2026},
url = {https://forkoff.xyz/research/asvc-discovery-gap-2026},
note = {4 LLMs, 48 datapoints, 14 named sources, FORKOFF ASVC methodology}
}First-party FORKOFF data leads; the aggregation draws on the strongest published research on AI citations. Each source below is named with its study and year, and linked.
| Source | Study and sample | Year |
|---|---|---|
| FORKOFF Research | The Discovery Gap, 12 paired sets x 4 engines = 48 datapoints | 2026 |
| FORKOFF Research | GEO Citation Lab Rerun, 50 prompts x 5 surfaces = 250 measurements | 2026 |
| Seer Interactive | AIO Impact on Google CTR 2026 Update, 53 brands / 5.47M queries / 2.43B impressions | 2026 |
| Semrush | The Ghost Citations Study, 3,981 appearances / 115 prompts / 14 countries / 4 engines | 2026 |
| Semrush | AI Search Visibility Study, 150,000 citations / 5,000 keywords | 2025 |
| Profound | AI Platform Citation Patterns, 680M citations | 2025 |
| Ahrefs | Top Brand Visibility Factors, 75,000 brands studied | 2025 |
| Ahrefs | 38% of AI Overview Citations Pull From The Top 10, 863K SERPs / 4M AIO URLs | 2026 |
| Ahrefs | AI Search Overlap Study, AI-cited URL vs top-10 rank | 2025 |
| Ahrefs | 90+ AI SEO Statistics, first-party clickstream + index | 2025 |
| Ahrefs | AI Overview Brand Correlation, mention-quartile multiplier | 2025 |
| AuthorityTech | AI Citation 11% Platform Overlap Audit, 21,143 citations | 2026 |
| Pew Research Center | Google AI summary click study, 900 US adults + browsing data | 2025 |
| Princeton (Aggarwal et al.) | GEO: Generative Engine Optimization, GEO-bench (KDD) | 2024 |
| Search Engine Land | AI search engines cite Reddit, YouTube, LinkedIn most | 2025 |
FORKOFF built the AI Search Visibility Checker because the gap is the structural problem we solve. Run it on your own brand for the per-engine gap score. Pair the output with the cold email open rate benchmarks if you want to align outbound with the discovery layer. FORKOFF runs the three-lever fix as an engagement when the gap is wider than 60 points. Owning the methodology is the authority claim; the audit number is the call to action.
Authorship
Kshitij JK
Founder, FORKOFF
Reviewed by: Kartik Chugh
Last reviewed:
Published:
Methodology
The ASVC (AI Search Visibility Coverage) discovery-gap dataset is generated by running 12 paired branded-versus-head-term query sets against ChatGPT, Claude, Perplexity, and Gemini (48 datapoints per audit), then mapping cited URLs to domains and computing per-domain coverage. The surrounding aggregation adds statistics from 14 named authorities, each traced to a primary named source with a report title and year; untraceable figures were excluded. Methodology mirrors Princeton's GEO citation-analysis framework (KDD 2024) extended with FORKOFF's domain-disambiguation step that filters out parasite-SEO false-positive hits.
Sources cited
Have a question about this research methodology, or need help calibrating against your campaign data? Book a 30-min strategist call
AI Overview optimization is structural, not a domain-rating game. The 12 on-page patterns that make a page extractable and citable, with first-party data.
Read
DeFi protocol marketing is a TVL problem. The 2026 playbook to take a protocol from zero to its first real Total Value Locked, backed by on-chain data.
Read
DePIN marketing is a distribution and trust problem, not a hardware one. The 2026 playbook to take a network from testnet to a token it keeps using.
Read