Direct Answer
AI visibility is the measurable presence and representation of a brand, product, person, or source inside AI-generated answers. A brand can be named and linked, named without a link, cited as a source without being named, retrieved without a visible citation, or absent. Each state creates a different outcome: awareness, referral traffic, factual authority, research evidence, or a clear visibility gap.
In lesson 1.3 of the Ahrefs AEO course, Sam Oh introduces three practical states: cited and linked, mentioned but not linked, and not visible. That is an accessible starting point. For measurement, however, mentions and citations work better as two independent dimensions. An answer may cite a company page without naming the company, or name a company while linking only to third-party evidence.
Watch the Ahrefs Lesson
Video and course credit: Sam Oh and Ahrefs. The official lesson page lists a 6:27 runtime. This article preserves Oh's framework, checks the statistics against their original Ahrefs studies, updates the current Brand Radar metric definitions, and adds a no-click measurement system. Ahrefs' studies are observational and tool-specific; they do not reveal proprietary model or ranking algorithms.
The Three Types of AI Visibility
| Course state | What the user sees | Primary value | Main limitation |
|---|---|---|---|
| 1. Cited and linked | The answer names the brand or uses its material and includes a clickable source. | Referral traffic, attribution, verification, and possible conversion. | The link may support only one factual claim and may not function as an endorsement. |
| 2. Mentioned, not linked | The answer names or recommends the brand without a clickable owned URL. | Awareness, shortlist entry, branded search, and message association. | Direct attribution is weak and the claim may be supported by a third party. |
| 3. Not visible | The topic is discussed, but the brand is absent from the answer and sources. | A diagnostic gap that can guide research and prioritization. | Absence in one run is not proof of permanent invisibility. |
The third state deserves precision. If an AI answer recommends the category, workflow, or criteria but never names the brand, the business is not receiving brand visibility. It is receiving evidence that relevant demand exists. That is an opportunity signal, not a soft win.
A Better Model: The Mention-Citation Matrix
A mention is an entity event: the answer names the brand. A citation is a source event: the answer links to a page. They overlap, but neither guarantees the other.
| Answer state | Brand named? | Owned page cited? | What to measure |
|---|---|---|---|
| Named and cited | Yes | Yes | Recommendation accuracy, citation support, referrals, and conversions. |
| Named, not cited | Yes | No | Sentiment, category association, branded search, direct traffic, and self-reported attribution. |
| Cited, not named | No | Yes | Source authority, referred sessions, cited page, claim alignment, and assisted outcomes. |
| Neither | No | No | Competitor presence, cited evidence, missing qualifications, and opportunity value. |
There is a fifth internal state: found but not cited. Ahrefs' current Brand Radar documentation says some pages can be retrieved in the background yet omitted from the visible source list. It groups cited and background-retrieved pages under "Found in." This can help diagnose retrieval, but it is not user-visible brand awareness and should not be reported as a citation.
What the 28% Citation Study Really Measured
The course says only about 28% of AI mentions include a link. The original November 2025 Ahrefs study is more specific: it analyzed more than 31,000 mentions of Ahrefs across six AI platforms in Brand Radar's 150-million-prompt database. It measured whether a mention of the Ahrefs brand included a citation to the Ahrefs domain.
| Platform | Ahrefs mentions with an Ahrefs link | What the number does not prove |
|---|---|---|
| Google AI Overviews | 10.7% | A universal citation rate for all sites in AI Overviews. |
| Gemini | 16.8% | That Gemini rarely cites sources overall. |
| Microsoft Copilot | 26.1% | How often other brands or source types receive links. |
| ChatGPT | 26.9% | A fixed probability that a ChatGPT brand mention links to the brand. |
| Google AI Mode | 36.8% | That AI Mode always links more than every other Google surface. |
| Perplexity | 51.6% | That every Perplexity mention or answer contains an owned-brand citation. |
The percentages are valuable because they demonstrate how different mention and citation behavior can be. They are not benchmarks a company should promise to match. Brand size, query mix, domain setup, industry, country, platform version, and study date all affect the result.
Linked Mentions Had More Estimated Exposure, With a Catch
When Ahrefs weighted the same mentions by its impression estimate, linked mentions accounted for 78.4% of Perplexity exposure, 71.3% of Gemini, 46.2% of AI Mode, 36.6% of ChatGPT, 31.7% of Copilot, and 13.0% of AI Overview exposure. That is why the lesson says links tended to appear on higher-traffic queries.
Ahrefs also states the critical limitation: its SEO strategy deliberately targets high-volume keywords, so the study cannot establish that AI platforms naturally cite more often on popular queries. It shows that Ahrefs' linked mentions occurred on prompts associated with greater Google search demand.
Do Unlinked Mentions Compound Into Future AI Visibility?
The transcript explains unlinked mentions as training examples that strengthen the association between a brand and a topic. This is plausible as a broad mental model, but it is too causal when applied to one article, review, or video.
A public mention may enter a model's training data, be retrieved at answer time, appear through a licensed corpus, remain uncrawled, or be excluded by provider and publisher controls. Training and live retrieval are separate paths. OpenAI, for example, documents different controls for OAI-SearchBot and GPTBot. A company cannot verify that one new web mention changed model weights or caused a future answer.
What the evidence supports is correlation. Ahrefs' original 75,000-brand study found a 0.664 Spearman correlation between branded web mentions and appearance in AI Overviews, stronger than the link and DR metrics it tested. Its later cross-platform update found branded web mentions correlated at 0.656 for ChatGPT, 0.709 for AI Mode, and 0.664 for AI Overviews. YouTube mentions correlated even more strongly, roughly 0.71 to 0.74 across those platforms.
This does not prove that mentions cause visibility. Famous, useful, well-distributed brands naturally earn more coverage, branded searches, links, reviews, and AI mentions at the same time. The actionable conclusion is still sensible: earn accurate, relevant mentions in places your audience trusts. Do it for human reach and independent corroboration first, with AI visibility as an additional benefit.
The Response Type Changes the Value
| Response type | Useful visibility | Best evidence asset | Likely business outcome |
|---|---|---|---|
| Product shortlist or comparison | Named recommendation plus product or category-page citation. | Clear positioning, comparison criteria, pricing, reviews, and third-party proof. | Shortlist entry, qualified referral, or purchase. |
| How-to guide | Expert mention, procedural citation, or embedded video. | Tested steps, screenshots, examples, failure modes, and dated instructions. | Trust, tool adoption, newsletter signup, or service inquiry. |
| Direct factual answer | Source citation or accurate entity association. | Primary data, definitions, documentation, and methodology. | Authority and awareness, often with low click intent. |
| Local or professional recommendation | Named recommendation with location, fit, and qualification. | Consistent business details, reviews, credentials, service scope, and local proof. | Call, visit, consultation, or branded search. |
| Video answer | Video citation, creator mention, or embedded result. | Searchable title, spoken evidence, chapters, description, and a strong visual explanation. | Watch time, subscriber growth, and brand familiarity. |
A citation is most commercially valuable when it appears inside a decision the business can serve. A million estimated impressions on a definition may matter less than 500 appearances in high-intent comparisons. Weight visibility by decision value, factual accuracy, and conversion path, not volume alone.
What the Brand Radar Metrics Mean in 2026
| Metric | Current Ahrefs definition | Do not interpret it as |
|---|---|---|
| Mention | One count when a brand appears at least once in an AI response, even if named several times. | One count per textual occurrence or proof of positive sentiment. |
| Citation | One count when a page appears at least once as a cited source in an AI response; domains are counted once per response when any page is cited. | A brand endorsement, click, or conversion. |
| Found in | A page was cited or retrieved in the background while the answer was generated. | A visible citation. |
| Impressions | Estimated exposure based on Google search-volume proxies for prompts where the brand appears. | Actual platform impressions or unique users. |
| AI Share of Voice | The brand's share of estimated impressions compared with selected tracked brands. | A fixed market-share statistic independent of the competitor setup. |
The definitions have operational consequences. If the brand is named three times in one answer, Brand Radar still counts one mention. If five pages from one domain are cited, the domain count can still be one for that response. Share of Voice can move simply because the comparison set or entity configuration changed. Store the query, platform, entities, competitor list, country, and date range with every exported baseline.
Use a Layered Measurement Stack
Layer 1: answer visibility
Track repeated prompts, named mentions, sentiment, recommendation position, competitors, cited pages, supporting sources, and factual errors. Test the same prompt several times because answers and citations vary.
Layer 2: platform exposure
Use prompt indexes and modeled impressions for directional comparison. Keep the methodology visible. In June 2026, Google announced dedicated generative AI performance reports in Search Console for a subset of sites, covering Search features such as AI Overviews and AI Mode. Where available, those first-party reports are stronger evidence than third-party estimates for Google exposure.
Layer 3: referral behavior
Segment AI referrers in analytics and inspect landing pages, engagement, assisted conversions, and revenue. OpenAI documents utm_source=chatgpt.com on ChatGPT search referrals. Keep direct and unattributed traffic in view because users may search the brand later instead of clicking.
Layer 4: no-click demand
Monitor branded search, direct visits, social profile visits, app-store searches, and sales-call language. Add a short "How did you hear about us?" question with an AI-assistant option and free text. Treat this as supporting attribution, not perfect causality.
Layer 5: business outcome
Connect qualified leads, trials, purchases, retention, and support deflection to the decision topic. A mention is valuable only when the representation is accurate and the audience can take a useful next step.
A Practical Brand Radar Workflow
- Define the entity. Add the canonical brand name, common variants, products, and domain scope. Exclude ambiguous names that create false positives.
- Choose real competitors. Keep the comparison set stable and specific to one category or decision journey.
- Record the baseline. Export mentions, citations, impressions, Share of Voice, cited pages, and cited domains by platform.
- Calculate the mention-citation gap. Find responses that name the brand but do not cite an owned page, then inspect which sources support the mention.
- Calculate the demand-mention gap. Find high-value topic responses where competitors appear and the brand does not.
- Review accuracy. Mark wrong positioning, obsolete facts, unsupported praise, missing limitations, and competitor confusion.
- Map the evidence gap. Identify the owned page, independent source, video, documentation, or proof that would make the answer more useful.
- Retest monthly. Preserve the same settings, annotate campaigns and page changes, and compare distributions rather than one favorable answer.
The gap between modeled impressions and mentions is useful only after the prompt set is commercially relevant. A brand does not need to appear in every answer about its broad industry. Prioritize prompts where the business is genuinely qualified, the customer is making a decision, and the missing evidence can be improved honestly.
Which Visibility Type Matters for Your Business?
| Business model | Primary visibility target | Secondary signal |
|---|---|---|
| Ecommerce or SaaS | Named product recommendation with a relevant product, pricing, comparison, or review citation. | Qualified referrals, trial starts, branded search, and assisted revenue. |
| Local service | Accurate named recommendation with location, service fit, credentials, and trust evidence. | Calls, map actions, consultation forms, and "heard via AI" responses. |
| Publisher | Source citations on factual and explanatory queries. | Referral engagement, subscriptions, recurring citation coverage, and update demand. |
| Consultant or agency | Expert mention on high-value problems plus proof-page citations. | Branded search, direct inquiries, and sales-call attribution. |
| Creator | Named expertise and video citations in how-to, review, and comparison answers. | Watch time, subscribers, profile visits, and sponsorship interest. |
| Research or documentation team | Accurate citations to primary evidence, even without a prominent brand mention. | Dataset reuse, backlinks, policy references, and expert recognition. |
A 30-Day AI Visibility Audit
Week 1: define and sample
Select one decision journey and 15 prompts. Run each prompt three to five times on two relevant platforms. Record mentions, citations, recommendation context, cited pages, competitors, and factual errors.
Week 2: map evidence
Separate named-and-cited, named-only, cited-only, and absent responses. For the most valuable gaps, map the best owned page and credible external sources. Check crawler access, canonical URLs, entity names, and outdated claims.
Week 3: improve one source and one corroboration path
Strengthen one owned page with a direct answer, primary evidence, clear limitations, current dates, and an obvious next step. Earn or publish one genuinely useful independent mention, demonstration, review, expert contribution, or video. Do not seed fake community consensus.
Week 4: retest and report
Repeat the fixed tests under the same conditions. Report change in answer distribution, factual accuracy, modeled exposure, referral sessions, and business outcomes. Mark any improvement as observational unless the experiment isolates a cause.
Copy-Ready AI Visibility Worksheet
Objective: measure one brand's visibility for one customer decision.
Brand and domain: [canonical entity / domain]
Category and audience: [specific market]
Decision journey: [research / shortlist / purchase / implementation]
Platforms: [two surfaces]
Competitors: [stable comparison set]
Test dates: [range]
For every prompt and run, record:
- platform, product/model, country, login state, conversation context
- brand mentioned: yes / no
- recommendation accurate: yes / partial / no
- owned page cited: yes / no
- third-party source cited: [domain and URL]
- owned page found but not cited: yes / no / unknown
- competitor mentions and citations
- material errors, missing qualifications, or outdated claims
- response type: comparison / how-to / direct answer / local / video
- buyer value: low / medium / high
Report these separately:
1. Named + cited responses.
2. Named without owned citation.
3. Owned citation without brand mention.
4. Neither mentioned nor cited.
5. Factual accuracy and sentiment.
6. Modeled impressions and AI Share of Voice, with methodology.
7. Referral sessions and assisted conversions.
8. Branded search, direct traffic, and self-reported attribution.
30-day actions:
- fix one entity, crawl, or canonical problem
- improve one owned evidence page
- earn one relevant independent mention or demonstration
- retest the identical prompt set
- connect visibility changes to a business outcome
Rules:
- Do not call modeled impressions actual views.
- Do not treat correlation as proof of causation.
- Do not report background retrieval as a visible citation.
- Do not infer sentiment from mention count alone.
- Do not compare Share of Voice with different competitor sets.Video Chapters
| Time | Topic |
|---|---|
| 00:00 | What winning AI search means |
| 00:42 | The three types of AI visibility |
| 00:47 | Cited and linked |
| 01:01 | Mentioned but not linked |
| 01:21 | Not visible at all |
| 01:35 | How often Ahrefs mentions included links |
| 02:21 | Weighting linked mentions by impressions |
| 03:01 | The case for unlinked brand mentions |
| 03:36 | The 75,000-brand correlation study |
| 04:04 | Step-by-step, direct-answer, and video responses |
| 04:58 | AI visibility as a spectrum |
| 05:07 | Visibility priorities by business model |
| 05:47 | Measuring with Ahrefs Brand Radar |
| 06:08 | The impressions-to-mentions opportunity gap |
Bottom Line
Sam Oh's three-state model makes the essential point: AI visibility is wider than referral traffic. A linked citation is measurable and useful, but an accurate unlinked recommendation can create awareness and branded demand. An owned page can also be cited without the publisher becoming memorable. These outcomes should not be collapsed into one score.
The practical system is a mention-citation matrix connected to real outcomes. Track whether the brand is named, whether an owned page is cited, how the answer frames the company, what evidence supports it, and what the user does next. Use modeled impressions and Share of Voice directionally, preserve the method, and never turn correlation into a promise. The goal is not to maximize appearances. It is to become an accurate, credible, useful part of the decisions the business is qualified to serve.
Sources and Link Map
- Ahrefs video: What Is AI Visibility? The 3 Types Every Marketer Needs to Know - the embedded source lesson.
- Ahrefs Academy lesson 1.3 - official lesson page and 6:27 runtime.
- Ahrefs: Do AI Assistants Link When Mentioning Brands? - the 31,000-mention Ahrefs case study, platform citation rates, impression-weighted results, and methodology caveat.
- Ahrefs Help Center: AI Visibility Metrics - current definitions of mentions, citations, Found in, impressions, and AI Share of Voice.
- Ahrefs: Top Brand Visibility Factors - updated 75,000-brand correlations across ChatGPT, AI Mode, and AI Overviews.
- Ahrefs: AI Visibility Audit - Brand Radar setup, competitor comparison, topic gaps, and cited-page analysis.
- Ahrefs: Monitor and Win Brand Mentions in AI Answers - monitoring cadence, third-party sources, and content opportunities.
- Google Search Central: generative AI performance reports - June 2026 announcement of dedicated Search Console views for a subset of sites.
- OpenAI: Publishers and Developers FAQ - OAI-SearchBot, GPTBot, public-site eligibility, and ChatGPT referral tracking.
- Perplexity: How does Perplexity work? - official description of real-time search and visible source citations.
- JQ AI SYSTEMS: Google AI Overviews vs ChatGPT vs Perplexity - platform retrieval, source-mix, and prioritization differences.
- JQ AI SYSTEMS: How AI Search Engines Work - retrieval, query fan-out, evidence selection, and citation mechanics.
- JQ AI SYSTEMS: What Is AEO? - terminology, course map, and a 30-day baseline.