Direct Answer
The useful 80% of AEO is a five-part operating system: understand how AI search expands a question, map where your brand is absent, research the prompts and intents that matter, publish and earn evidence worth citing, then measure visibility through to revenue. Ahrefs explains that system clearly in 19 minutes. The framework is strong. Some statistics and crawler advice need an August 2026 update.
The central idea survives every update: do not optimize for one exact keyword or one dashboard score. Build a topic-level evidence footprint across your website, trusted third-party pages, communities, and video. Make the important facts consistent. Give search systems a crawlable, direct answer. Then judge the work by mentions, citations, qualified visits, branded demand, and conversions together.
llms.txt trick, or a guaranteed way to make an assistant recommend a brand.
Watch the Ahrefs AEO Guide
Video credit: Ahrefs. Watch the original video on YouTube. Ahrefs' studies are valuable observational evidence from its own indexes and products. Correlations do not establish ranking factors or causation.
Four Important Updates to the Video
| Video takeaway | Current evidence | How to act |
|---|---|---|
| 76% of AI Overview citations rank in Google's top 10 | Ahrefs' March 2026 update found 37.1% in the top 10 across four million cited URLs | Keep SEO strong, but research the related fan-out queries and pages beyond the original SERP. |
| Check GPTBot, ClaudeBot, and Google-Extended for visibility | Those names primarily concern training or other Gemini systems. Search crawlers and Googlebot have different roles. | Govern each crawler by purpose instead of allowing or blocking every AI bot together. |
| Atomic writing and BLUF help AI reuse content | They improve human clarity, but Google says no special chunking or AI-specific writing format is required. | Use direct sections because they serve readers and quotability, not because they are confirmed ranking factors. |
| AI visibility is mostly hidden | Google began testing dedicated generative-AI Search Console reports in June 2026 for a subset of sites. | Use the new report when available, while retaining referral, bot, brand-demand, and conversion measurement. |
None of these corrections invalidates the Ahrefs framework. They make it more precise. AI search changes quickly enough that every statistic needs a date, platform, country, dataset, and method attached.
1. Understand Query Fan-Out
Google officially describes query fan-out as a set of concurrent related searches generated to answer one broader question. A request such as "plan a five-day Japan trip in November" can trigger searches about Tokyo neighborhoods, Kyoto weather, rail passes, budgets, and seasonal events. The final answer can cite pages that rank for those subtopics even when they do not rank for the original prompt.
This changes the unit of planning from one keyword to one decision journey. A cybersecurity consultancy should not only target "cybersecurity consultant." It needs evidence for questions about incident readiness, vendor risk, employee training, audit scope, response time, pricing, and industry obligations. One comprehensive hub can answer several of those needs, supported by focused pages where the intent genuinely differs.
A practical fan-out map
| Layer | Question to map | Best evidence |
|---|---|---|
| Problem | What is going wrong, and how is it diagnosed? | Diagnostic guide, checklist, original data |
| Options | What approaches or products could solve it? | Comparison with explicit criteria |
| Fit | Who is each option for or not for? | Use cases, limitations, decision tree |
| Risk | What can fail, cost more, or create liability? | Failure modes, controls, expert review |
| Proof | Has this worked in a comparable situation? | Case study, demonstration, third-party reference |
| Action | What should the buyer do next? | Implementation plan, calculator, assessment |
2. Map Mentions, Citations, Competitors, and Topic Gaps
Ahrefs demonstrates a brand-gap analysis in Brand Radar. The tool can compare mentions, citations, estimated impressions, and AI share of voice across platforms. The method does not require Ahrefs, but it does require a fixed prompt set and consistent definitions.
| Gap | What it means | Likely response |
|---|---|---|
| Mention without citation | The assistant knows the brand but does not use its site as evidence | Publish a stronger source page, improve indexability, and earn links to the underlying evidence. |
| Competitor mention gap | A competitor appears for a relevant prompt and your brand does not | Study the cited sources and the buyer criteria they satisfy. Do not copy the page blindly. |
| Topic association gap | The brand is visible generally but absent from a topic it should own | Create a focused topic hub, expert proof, and consistent off-site references. |
| Platform gap | The brand appears in one assistant but not another | Compare the platform's source set, index, geography, and crawler access. |
| Accuracy gap | The brand is mentioned with wrong facts or positioning | Correct owned facts, structured data, profiles, listings, and authoritative third-party descriptions. |
Run the same prompts at least three times and record the model, mode, date, language, and location. Generated answers vary. A single screenshot is not a baseline. For the full measurement model, see why crawl access, mentions, citations, clicks, and conversions must stay separate.
3. Research Prompts, Intent, and the Zero-Click Outcome
Traditional keyword data is still useful because demand, language, difficulty, and business value do not disappear when the interface changes. Ahrefs suggests combining seed keywords with modifiers such as "best," "versus," "alternative," "review," "for," and "how." The result should become a prompt portfolio, not an indiscriminate list of pages to publish.
Then apply the video's "AI filter": could the generated answer satisfy the searcher without a click? Ahrefs' 146-million-SERP study found AI Overviews on about 20.5% of the measured keywords and 57.9% of question queries; 99.9% of triggered keywords carried informational intent in its classification. These are dataset observations, not permanent platform rules.
A low-click query is not automatically worthless. Change the objective. For an informational question, the win may be an accurate brand mention, cited proof, remembered category association, or a later branded search. For high-intent comparisons and implementation questions, a click and conversion remain realistic goals. Score each prompt against four outcomes: mention, citation, visit, conversion.
4. Earn Citations Across Four Surfaces
Owned content: answer well, not long
Ahrefs analyzed 174,048 pages cited in AI Overviews and found a 0.04 Spearman correlation between word count and citation position, effectively no useful relationship. About 53.4% of cited pages were under 1,000 words. The practical rule is simple: use the length the reader and evidence require. Do not pad a 500-word answer into a 3,000-word article.
Use a direct answer near the top, descriptive headings, one coherent question per section, named entities, dates, limits, and source links. "Bottom line up front" and self-contained sections make material easier to scan and quote. They are editorial practices, not confirmed AEO ranking factors. Google's current guidance explicitly says there is no required chunk size or special AI writing format.
Freshness also needs judgment. Ahrefs reports that AI-cited pages were 25.7% fresher than organic results across its 17-million-citation study. That does not justify changing a date without changing the substance. Refresh a page when prices, products, laws, interfaces, evidence, or recommendations have changed, and show readers what was reviewed.
Third-party mentions: build corroboration
Ahrefs' study of 75,000 brands found branded web mentions correlated with AI visibility at roughly 0.66 to 0.71 across the studied systems. This is an association, not proof that purchasing mentions causes recommendations. Strong brands naturally earn more coverage and more searches.
The defensible play is to create reasons for independent people to mention you: original research, a useful tool, a public template, a clear expert opinion, a documented result, or a product worth comparing. Contribute to Reddit, Quora, and specialist communities as a participant, not as a disguised campaign. Google specifically warns against inauthentic mentions. Our companion guide explains how to sequence owned content and third-party proof.
YouTube: a discovery and evidence surface
In Ahrefs' brand study, YouTube mentions had the strongest reported correlation with AI visibility, around 0.737. That is a reason to test video, not a license to call YouTube a ranking factor. The platform provides searchable titles, descriptions, chapters, transcripts, demonstrations, and third-party discussion that several search systems can retrieve.
- Choose an evergreen question with real demand. Tutorials, comparisons, setup guides, tests, and troubleshooting tend to have durable search intent.
- Name the topic clearly. Put the actual subject in the title and opening description rather than relying on a clever hook alone.
- Say the important entities aloud. Accurate transcripts should contain product names, versions, methods, and conclusions.
- Add chapters. Timestamps help viewers and create a structured map of the discussion.
- Show proof. Demonstrations, measurements, before-and-after evidence, and limitations make the video useful beyond its metadata.
- Publish a companion page. Embed the video with a unique written analysis, source links, and reusable evidence rather than a transcript dump.
Technical eligibility: keep the important evidence reachable
Google says pages need no special AEO schema or machine-readable file to appear in AI Overviews or AI Mode. They must be indexed, eligible for a normal Search snippet, crawlable, internally discoverable, and useful. Structured data should match visible content because it supports normal Search understanding and rich results, not because it guarantees an AI citation.
Crawler Controls: Search Is Not Training
| Crawler or control | Primary purpose | Visibility implication |
|---|---|---|
OAI-SearchBot | OpenAI search discovery | OpenAI says allowing it is important for inclusion in ChatGPT summaries and snippets. |
GPTBot | OpenAI model training | Blocking it is a training choice, not the same as blocking ChatGPT Search. |
Claude-SearchBot | Anthropic search indexing | Anthropic says disabling it may reduce visibility and accuracy in user search results. |
ClaudeBot | Anthropic model-development crawling | Can be governed separately from search and user-directed retrieval. |
Claude-User | User-directed page retrieval | Supports fetching content when a Claude user asks for it. |
Googlebot | Google Search crawling | Controls eligibility for Search, including supporting links in AI Overviews and AI Mode. |
Google-Extended | Some Gemini training and grounding uses | Google says it does not affect inclusion or ranking in Google Search. |
Check more than /robots.txt. A CDN, web application firewall, bot product, login wall, or JavaScript challenge can block a crawler even when robots rules allow it. Cloudflare now classifies Search, Agent, and Training behaviors separately. Its announced defaults for new domains change on 15 September 2026, after this article's publication date, so inspect the live dashboard rather than assuming one universal default.
robots.txt is advisory. It expresses policy to compliant crawlers; it is not a security boundary. Keep private or licensed content behind authentication and enforce access server-side.
5. Measure the Whole Journey
An AEO dashboard should answer six different questions. Combining them into one "visibility score" hides where the system is failing.
| Layer | Metric | What it can tell you | What it cannot prove |
|---|---|---|---|
| Access | Verified crawler requests and status codes | Whether a bot reached a page | That the page was indexed, cited, or trusted |
| Google visibility | Generative-AI Search Console impressions and pages, where available | Which pages appeared in Google's AI features | Full query detail or business influence |
| Answer visibility | Mentions, citations, and share of voice on a fixed prompt set | Whether the brand appears and which sources support it | A stable ranking or causal influence |
| Demand | Branded searches, direct visits, branded impressions | Whether awareness may be growing | Which AI answer caused the change |
| Traffic | AI referrals and landing pages | Directional click-through from assistants | All AI-influenced visits, because referrers can be missing |
| Business | Qualified leads, revenue, and "How did you hear about us?" | Whether AI influenced a valuable outcome | Perfect attribution without supporting context |
Google launched dedicated generative-AI performance reports in Search Console in June 2026 for a subset of sites. The reports expose impressions, pages, countries, devices, and dates for AI features, while the data remains included in overall Search performance. Availability is still limited, so do not promise every client a report they may not have.
Ahrefs' often-cited conversion example is specific to Ahrefs: during one 30-day period, AI search represented 0.5% of visits and 12.1% of signups, a 23-times higher conversion-to-visit ratio than organic search. Treat it as a case study, not an industry benchmark. The video's recommendation to collect self-reported attribution is the durable lesson because AI-assisted journeys often end in a later branded or direct visit.
A 30-Day AEO Plan for a Small Team
- Days 1-3: establish the baseline. Choose 20 buyer prompts, three competitors, two markets or languages, and two AI systems. Record mentions, citations, source domains, and factual errors across three runs.
- Days 4-5: audit access by purpose. Check Googlebot, OAI-SearchBot, Claude-SearchBot, firewall events, status codes, canonical tags, snippet controls, and indexability. Make a separate decision about training crawlers.
- Days 6-10: map the decision journey. Turn the 20 prompts into problem, options, fit, risk, proof, and action clusters. Merge overlapping intent.
- Days 11-16: improve five source pages. Add a direct answer, first-hand evidence, dates, limitations, named entities, source links, and an appropriate next action.
- Days 17-21: create one proof asset. Publish a small study, calculator, checklist, demonstration, template, or case study that other people have a reason to reference.
- Days 22-25: distribute honestly. Pitch relevant publications, answer real community questions, and create one searchable YouTube video or creator collaboration.
- Days 26-30: measure and decide. Re-run the prompt set. Compare citations, mention accuracy, Search Console visibility, AI referrals, branded demand, and qualified conversions. Choose one next experiment.
Three anti-patterns to avoid
- Prompt-page multiplication: dozens of near-duplicate pages aimed at imagined fan-out queries.
- Manufactured consensus: fake Reddit participation, paid undisclosed mentions, or self-promotional "best" lists presented as independent evidence.
- Metric theater: reporting bot hits or a tool's share-of-voice score as if they were revenue.
Continue Learning With Ahrefs
The 19-minute video is the fast overview. These three Ahrefs resources extend the same system without forcing three more video players into the page:
- AI SEO Course for Beginners: Complete AEO Tutorial - a longer end-to-end course for implementing the workflow.
- Creators Built the Web. Google AI Is Rewriting It. - useful context on how generated answers change discovery and publisher incentives.
- Top SEO Experts Build Me an AI Search Strategy - an expert-led strategy session for comparing approaches and tradeoffs.
Video Chapters
| Time | Topic | Practical takeaway |
|---|---|---|
| 00:00 | Why some brands get recommended | AEO works as a repeatable system, not luck. |
| 00:40 | How AI search works | Plan for topic-level fan-out, not one head term. |
| 02:46 | Brand gap analysis | Separate missed citations, competitor gaps, and topic gaps. |
| 03:54 | Keyword and prompt research | Use familiar demand data, then evaluate the likely zero-click outcome. |
| 06:38 | Content and citations | Answer directly, stay current, and match the format to the decision. |
| 09:22 | Off-site brand mentions | Build real corroboration across editorial, community, and owned surfaces. |
| 11:25 | YouTube visibility | Create searchable, explicit, transcript-rich demonstrations. |
| 13:28 | Technical access | Check robots and infrastructure without confusing search with training. |
| 14:34 | Measurement | Combine referrals, bot logs, prompts, brand demand, and self-reporting. |
| 17:21 | Custom AEO strategy | Fix the largest evidenced gap instead of copying a generic checklist. |
Bottom Line
Ahrefs compresses AEO well: mechanics, gaps, prompt research, citation surfaces, and measurement. The framework becomes more useful when its statistics are dated and its heuristics are kept separate from official requirements. The newer evidence says Google top-10 overlap is lower than the video's 76% figure, crawler purposes must be separated, and dedicated Google AI visibility reporting is beginning to emerge.
The durable strategy is pleasantly unglamorous. Build a crawlable site. Answer valuable questions with original evidence. Earn honest discussion beyond your own domain. Make video when it adds proof. Keep brand facts consistent. Measure the complete journey. That is most of AEO, and it is also the part least likely to disappear with the next model update.
Sources and Further Reading
- Ahrefs: Learn 80% of AEO in 19 Minutes
- Google Search Central: optimizing for generative AI features
- Google Search Central: generative-AI performance reports
- Ahrefs: updated AI Overview and top-10 overlap study
- Ahrefs: 75,000-brand AI visibility correlation study
- Ahrefs: freshness and AI citations
- Ahrefs: word count across 174,048 cited pages
- Ahrefs: 146-million-SERP AI Overview trigger study
- Ahrefs: its AI referral and signup case study
- OpenAI: publisher and developer crawler guidance
- Anthropic: Claude crawler purposes and controls
- Google: Google-Extended and Search clarification
- Cloudflare: AI bot policies and September 2026 defaults