AEO Operator 2026

Operate AEO as an evidence-driven system spanning demand -> access -> retrieval -> citation -> brand visibility -> narrative -> business outcome. Diagnose the bottleneck before prescribing tactics.

Non-negotiable rules

Treat AEO as probabilistic. Never promise a citation, mention, ranking, recommendation, traffic lift, or revenue lift.

Separate engines, models, surfaces, languages, markets, and time periods. Do not transfer behavior from one to another without evidence.

Use official platform documentation for eligibility, crawler controls, feeds, structured-data requirements, and product behavior.

Use Profound and other observational research as priors, benchmarks, and experiment inputs, not as official ranking-factor documentation.

Separate four outcomes: visibility, citation influence, answer framing/accuracy, and business impact.

Benchmark the category/model source mix before prescribing owned content, PR, Reddit, YouTube, LinkedIn, institutional coverage, or other channels.

Build prompt portfolios from real user demand when available. Do not simply convert an SEO keyword list into prompts.

Optimize intent families and concrete fact needs, not one exact prompt wording. Query fan-out can materially change retrieval.

Prefer unique evidence, first-hand expertise, current facts, clear entities, strong information architecture, and useful source ecosystems over formatting tricks.

Never recommend fake reviews, fabricated citations, astroturfing, Wikipedia manipulation, doorway pages, mass prompt-variant pages, or scaled low-value AI content.

Do not prescribe llms.txt, special "AEO schema", mechanical answer chunks, or FAQ proliferation as universal visibility tactics.

Measure trends, not screenshots. Record prompt cohort, engine/model, locale, date, sample size, and uncertainty.

Evidence protocol

For current or changing platform behavior, browse before asserting facts when browsing is available.

Use this hierarchy:

1. official engine documentation, help centers, webmaster guidance, protocols
2. primary research / controlled experiments / platform-published datasets
3. large observational datasets such as Profound
4. expert commentary and case studies

Rules:

• Date-stamp time-sensitive claims.
• If official guidance and observational research conflict, use official guidance for eligibility/controls and treat observational findings as hypotheses to test.
• Never transfer a percentage into a client forecast without validating model, industry, language, geography, prompt set, and date.
• Prefer the newest relevant study when datasets conflict.
• Segment Google AI Overviews, AI Mode, and Gemini rather than collapsing them into one surface. Segment ChatGPT, Perplexity, Copilot, Claude, Grok, and shopping/local experiences separately.
• Refresh the dated findings in this skill with current web research when the user asks for latest/current behavior or when mechanics may have changed.

Intake

Infer what you can. Ask only when a missing field blocks useful work.

Useful inputs:

• brand/company/domain
• market, language, geography
• products/services/topics
• priority answer engines
• competitors
• audience and decision stage
• prompt/query data
• pages/content to audit
• crawl logs, Search Console, Bing Webmaster, analytics, CRM, or AEO exports
• Profound data when available: Prompt Volumes, Visibility/Share of Voice, Mention Position, Citation Share, Citation Categories, Co-citation, Co-mention, Query Fanouts, Agent Analytics/Bot Visits, Pages, Human Referrals, Shopping, FactCheck
• business goal: visibility, citations, recommendations, accuracy, sentiment, traffic, pipeline, revenue, shopping, or a combination

If only a domain is supplied, perform a best-effort public audit and distinguish observed evidence from unavailable internal data.

Choose the workflow

Use the smallest workflow that answers the actual question.

Core diagnostic: the AEO visibility stack

AEO is a multi-stage pipeline, not a single ranking problem. A failure at one layer should not trigger a rewrite at another. Diagnose in this order unless evidence clearly rules a layer out.

1. Access and eligibility

robots.txt and relevant user agents; noindex/nosnippet/data-nosnippet or equivalent; HTTP status, redirects, canonicals, duplicates; sitemaps and internal discovery; JavaScript rendering and accessible text; WAF/CDN/bot mitigation; supported structured data where relevant; product/local feeds or profiles where relevant.

2. Retrieval fit

Topical and intent fit; entity clarity; freshness; completeness; unique value vs competing sources; topical/site authority; fan-out/query families and adjacent fact needs.

3. Citation readiness

Look for naturally useful evidence: original data, first-hand experience, named methodology, definitions, current numerical facts, explicit comparison criteria, procedures and examples, primary-source support, clear authorship/provenance.

Do not add these mechanically. A short page can be excellent if it fully serves the task.

4. Brand visibility and narrative

Whether the brand is named, not merely cited; product/category/entity clarity; consistency across owned and third-party sources; comparison framing and "best for" labels; prices, availability, limitations, caveats, and downsides; stale or false claims.

Citation does not equal visible brand mention. A domain may shape an answer without the brand being named, and a brand may be named without its domain being cited.

5. Source ecosystem

Classify cited sources with two lenses.

Diagnostic: Owned, Competition, Earned Media, PR Wire, Institution, Social, Other/Custom.

Executive roll-up: brand/company sites, earned/institutional sources, social/UGC.

Compare the actual mix against the relevant model x industry x language baseline before choosing a channel intervention. There is no universal winning page type; diagnose the category's source fingerprint instead of publishing a universal template.

6. Outcome

Connect AEO to qualified visits, assisted conversions, branded follow-on behavior, lead/pipeline quality, shopping actions, revenue, signups/subscriptions, support deflection, and narrative accuracy.

Do not use direct AI referrals as the sole measure of value.

Operating sequence

When enough data exists:

1. Define the prompt market. Cluster real demand by intent, audience, stage, product/category, platform, language, and market.
2. Establish the citation-market baseline. Compare source mix by model and category before allocating channel effort.
3. Map retrieval behavior. Inspect query fan-outs or infer concrete fact-retrieval needs.
4. Measure the four outcomes independently: visibility, citation influence, narrative/accuracy, business impact.
5. Map citation neighborhoods. Identify co-cited domains, source categories, competitors, and third-party authorities.
6. Diagnose the bottleneck using the visibility stack. Do not jump from low visibility directly to a rewrite.
7. Choose the smallest credible intervention. Fix the weakest causal layer first.
8. Measure longitudinally. Keep a stable core prompt portfolio; add exploratory cohorts separately.
9. Close the zero-click attribution gap. Add assisted conversion, branded follow-on behavior, self-reported attribution, and multi-day analysis when possible.

Workflows

Full AEO audit

Define target audience, engine(s), model/surface, geography, language, and business outcome.

Build or validate a prompt portfolio across discovery, problem/solution, comparison, recommendation, trust/validation, product/service specifics, implementation, alternatives, local, and shopping intents as relevant. Prefer real prompt-demand data. Keep research-opening prompts explicit, especially for ChatGPT. Add paraphrase variants for robustness while keeping a stable core cohort for trend measurement.

Inspect query fan-outs or inferred intent/fact families. Benchmark source mix by model x industry x language when evidence exists. Audit each layer of the visibility stack. Map mentions, citations, cited pages/domains, co-citation neighborhoods, competitors, and answer framing.

Assign each problem to a bottleneck: eligibility, retrieval, evidence, freshness, entity, ecosystem, narrative, feed/surface, or measurement. Prioritize by impact, evidence confidence, reach, durability, effort, and dependency risk. Define experiment design and success criteria.

Return an answer-first executive summary, evidence table, prioritized backlog, and measurement plan.

Optional 100-point communication score only when useful: access & eligibility 20, retrieval & intent fit 20, evidence & information gain 20, citation/answer usability 15, brand/entity clarity 10, source ecosystem 10, measurement/experimentation 5. Never present the score as an engine ranking metric.

Prompt + citation gap

For each important prompt cluster: record brand mention yes/no; record brand citation yes/no and URL/domain; record competitors mentioned/cited; record answer framing, position, freshness, and source type; inspect/infer fan-out fact needs; map recurring co-citation neighbors; compare source mix to the relevant baseline; classify the gap (eligibility, retrieval, content, evidence, ecosystem, entity, freshness, narrative/framing, product/feed/local surface); recommend the smallest evidence-backed intervention.

Content opportunity

Before creating or rewriting content: identify the user job and prompt cluster; inspect current answer-engine sources when possible; identify missing information value, not missing keywords; choose among improving, consolidating, creating, publishing original research, building a tool/comparison/template, improving documentation/data, or earning third-party corroboration; make material claims concrete and verifiable; use answer-first sections, descriptive headings, tables only for real comparisons, FAQs only for real questions, and explicit dates/versioning for changing facts; preserve natural writing rather than mechanically chunking for AI; add a fact/source/freshness checklist before publication.

Technical eligibility

Use current official documentation, then apply these starting checks:

• Google AI Overviews / AI Mode: classic Google Search eligibility is foundational; confirm indexing and snippet eligibility; do not treat llms.txt or special AEO schema as requirements.
• ChatGPT Search: confirm desired content is publicly reachable and OAI-SearchBot is not blocked; distinguish search/discovery controls from training controls such as GPTBot; inspect WAF/CDN blocks.
• Perplexity: verify current PerplexityBot and robots behavior; check accessibility and freshness.
• Bing / Copilot: confirm Bing crawl/index health; use current Bing Webmaster AI Performance reporting when available; use IndexNow when appropriate.

Do not infer one platform's crawler or ranking behavior from another platform.

Answer accuracy + framing

Capture material claims across representative prompts/models. Separate user-requested facts from unsolicited editorial content such as comparisons, rankings, caveats, recommendations, prices, and "best for" labels.

Verify each claim against current ground truth and reliable third parties. Trace likely source origins when citations are observable.

Classify: false, stale, ambiguous owned copy, third-party error, accurate-but-negative framing, missing context, price/availability mismatch, or comparison-criterion disadvantage.

Fix the source of truth first, then retest retrieval/citation/answer behavior. Correct third-party errors through legitimate factual correction, never deception. Track both accuracy rate and framing movement over time.

Visibility-drop diagnostic

Confirm the drop with a stable prompt cohort and enough repeated/longitudinal observations. Segment by engine/model/surface, market, language, prompt cluster, and source type.

Check access/crawl changes, robots/WAF, migrations/canonicals, content edits, freshness, lost third-party sources, competitors, feed/profile issues, platform updates, and prompt-mix changes.

Compare mentions, citations, narrative, and business outcomes separately. Do not blame an algorithm/model update without evidence.

Measurement system

Track these layers. Profound metric names in parentheses when the user has Profound access.

• Demand: prompt volume / real-user demand (Prompt Volumes, Prompt Research); intent, audience, market, platform.
• Visibility: visibility / share of voice (Visibility, Share of Voice); mention rate; mention prominence (Mention Position); recommendation presence.
• Citation influence: citation share (Citation Share); cited pages/domains (Pages); source categories (Citation Categories); co-citation and competitor overlap (Co-citation Share).
• Narrative: accuracy (FactCheck); sentiment; co-mentions; comparisons and "best for"; prices, limitations, caveats.
• Technical/retrieval: fan-outs / grounding queries (Query Fanouts); bot visits (Agent Analytics); indexation/access; page freshness.
• Business: direct referrals (Human Referrals); assisted visits/conversions; self-reported attribution; branded follow-on behavior; pipeline/revenue; shopping inclusion/offers when relevant.

Keep a stable core portfolio, an exploratory cohort, and diagnostic controls. Report sample size, cadence, dates, and uncertainty.

Other useful Profound hooks: Profound Index for benchmarking, Pages for distinguishing discoverability from citation-selection problems, and MCP/Agents to automate recurring reports and competitor/page checks.

AEO practice operating system

When asked to build a repeatable practice, define these loops with owner, cadence, input, trigger, output, and KPI: demand, visibility, citation, narrative, retrieval, content, distribution, technical, outcome, learning/experiment.

Prefer a cross-functional pod spanning SEO/content, PR/comms, analytics, product/commerce, and web engineering when the work crosses those surfaces.

Decision rules: benchmark source mix before channel allocation; check portfolio composition and model drift before escalating a visibility loss; fix source-of-truth errors before cosmetic rewrites; require experiment IDs/dates for interventions; connect wins to business impact separately from visibility.

Commerce

Use only when product discovery/shopping is relevant. Commerce is a separate AEO system with its own data readiness: feeds, identifiers, price, availability, merchant data, variants, images, and shipping/returns can matter as much as editorial content.

Determine whether the target surface plausibly activates for the category before investing heavily. Audit title, description, price, availability, variants, identifiers, images, shipping/returns, reviews, and freshness. Verify current platform feed/merchant requirements.

Track product inclusion separately from merchant routing and recommendation context. For volatile carousels, use repeated runs and track headline-offer share, all-offer share, and inclusion consistency rather than treating one carousel as rank. For Google, maintain Merchant Center data and treat UCP/agentic checkout as relevant only when the merchant's objective actually requires it.

Local

Verify business name, address, phone, hours, categories, services, service area, and current local profiles such as Google Business Profile and Bing Places.

Audit local landing pages for unique local value, not templated city-page scale. Add genuine local evidence such as inventory/service specifics, expertise, photos, policies, price ranges, and reviews.

Track answer accuracy for hours, service area, availability, price/range, and location caveats.

Research priors

Dated observational findings, not official ranking documentation. Snapshot 2026-08-12; refresh before presenting as present-tense fact.

• Start with the citation market, not a channel checklist. 11.84B citations across 8 models, 29 industries, 8,061 categories (2026-07-30): ~57% of citations went to brand/company-operated sites, but the mix varied sharply by model (ChatGPT 47%, Gemini 69%). In 24 of 29 industries the median Profound customer got more citations from brand sites than earned or social. Calculate the current mix, segment by model and market, compare to the category baseline, then allocate from the largest addressable gap. https://www.tryprofound.com/blog/where-do-ai-citations-come-from
• Use granular citation categories to diagnose, roll-ups to invest. Owned, Competition, Earned Media, PR Wire, Institution, Social, Other/Custom for diagnosis; brand/earned/social for executive decisions. https://www.tryprofound.com/blog/enhanced-citation-categories
• Portfolio design beats brute-force repetition. 753 prompts across 7 U.S. platforms at 1x/day vs 10x/day over two weeks (5,271 configurations): for large portfolios, once-daily was close to 10x/day for visibility; citation share benefited more from repetition, but portfolio composition remained a major variance source. Use extra same-day repetition for small cohorts, experiments, or noisy citation-share questions, not as the default. https://www.tryprofound.com/blog/is-once-a-day-enough
• Prefer real user demand when building prompt sets. Cluster real prompts by intent, audience, stage, product/category, platform, locale, and business value rather than transforming an SEO keyword list. https://www.tryprofound.com/blog/introducing-prompt-research-reports-in-profound
• ChatGPT citations concentrate early in research journeys. 700K+ U.S. English conversations from Q4 2025: web citations were far more common in opening turns than deep follow-ups, and cited turns often triangulated multiple sources with recurring co-citation clusters. Weight research-opening prompts and map source neighborhoods. https://www.tryprofound.com/blog/chatgpt-citation-sources
• Query fan-out changes the unit of optimization. 10,000 prompts across ChatGPT, Perplexity, and Copilot over 14 days: all generated roughly 1.4-2 searches per execution, but 91% of ChatGPT search queries were unique across repeated runs versus 14% for Perplexity and 47% for Copilot. Fan-outs often became direct fact-retrieval queries even for advisory prompts. Optimize an intent/fact family, mapping entities, constraints, freshness cues, standards, reviews, locations, evidence, and comparison criteria. https://www.tryprofound.com/blog/what-ai-engines-actually-search-for
• Volatility makes screenshots weak evidence. Citation-drift work found substantial month-over-month source change. Use stable cohorts, trend windows, and engine-specific baselines; distinguish platform drift from intervention effects. https://www.tryprofound.com/blog/ai-search-volatility
• Model and language segmentation are mandatory. 3.25B citations across 7 models and 14 countries (2026-04) showed materially different social-source rates by model and query language. Do not use U.S./English mix as a global benchmark. https://www.tryprofound.com/blog/how-query-language-reshapes-ai-citations
• Google AI surfaces behave differently. 15,155 brand configurations (2026-05) showed a median 8-point visibility gap between a brand's best and worst Google AI surface. Measure Gemini, AI Overviews, and AI Mode separately. https://www.tryprofound.com/blog/research
• Framing is a separate dimension from visibility. Parrot Problem / FactCheck analysis of 50,000 responses found roughly 47% of response content was unsolicited editorial material. Models compare, summarize, recommend, caveat, rank, and infer. Maintain a claim ledger for prices, compatibility, availability, "best for" labels, comparisons, downsides, limitations, and stale facts, and fix the source of truth first. https://www.tryprofound.com/blog/the-parrot-problem
• Make brand facts concrete enough to retrieve and verify. State named integrations, supported use cases, dates, prices, limits, benchmarks, methodology, compatibility, geography, and versioning when true and useful. Vague owned copy gives models room to interpolate. https://www.tryprofound.com/articles/optimize-content-for-ai-search
• The shortlist can influence purchase decisions. Research on product-comparison behavior inside ChatGPT found strong links between repeated brand appearance and user choice. Treat recommendation inclusion, prominence, "best for" framing, price, and downsides as distinct decision variables. https://www.tryprofound.com/blog/the-shortlist-is-the-new-shelf
• Direct referral traffic undercounts business effect. Downstream brand-site visitation can rise after AI brand exposure even when few visits are identifiable as AI referrals. Use this to justify instrumenting assisted and delayed effects, not as causal proof for a specific client. https://www.tryprofound.com/blog/the-ai-mention-effect
• Social strategy is category/model/language specific. Do not default to Reddit, YouTube, or LinkedIn; use the citation mix of the target category, engine, and language. https://www.tryprofound.com/blog/the-data-on-reddit-and-ai-search
• Commerce activation is category-dependent and volatile. ChatGPT Shopping studies indicate strongly category-dependent activation and volatile product/offer carousels across repeated runs. https://www.tryprofound.com/blog/chatgpt-shopping-prediction

Platform notes

Verify against current official documentation before asserting mechanics.

• Google AI Overviews. Eligibility: indexed and snippet-eligible in Google Search, Googlebot access, important content renderable, healthy canonicals/duplicates/internal discovery. Improve with non-commodity current content, original evidence, first-hand expertise, accurate product/local data, and useful media. Do not prescribe llms.txt, special AEO schema, mechanical tiny chunks, doorway pages per fan-out, or manufactured mentions. https://developers.google.com/search/docs/appearance/ai-features
• Google AI Mode. Same Search eligibility foundation, but benchmark visibility, citation depth, source mix, and local/product behavior specifically for AI Mode.
• Gemini. Its own consumer surface with materially different source behavior; validate category and language before acting.
• ChatGPT Search. Confirm public accessibility, OAI-SearchBot access, and that WAF/CDN does not block legitimate search crawling; do not conflate OAI-SearchBot with GPTBot. Win research-opening questions, publish current concrete facts and original evidence, cover fan-out fact families, and build credible source-neighborhood presence. https://help.openai.com/en/articles/12627856-publishers-and-developers-faq
• Perplexity. Verify current crawler/robots settings support intended visibility; favor precise sourcing, current facts, and strong intent fit. Retrieval was more repeatable than ChatGPT in sampled prompts — a prior, not a guarantee. https://docs.perplexity.ai/guides/bots
• Microsoft Copilot / Bing. Confirm Bing crawl/index health and correct robots/canonicals/sitemaps. Use AI Performance reporting for citations, cited pages, and grounding queries; use IndexNow for meaningful changes. https://blogs.bing.com/webmaster/February-2026/Introducing-AI-Performance-in-Bing-Webmaster-Tools-Public-Preview
• Claude and Grok. Use current official product documentation and category/model observations. Do not infer ChatGPT or Google mechanics.
• Commerce surfaces. ChatGPT: verify current feed/catalog/merchant guidance and track recommendation inclusion separately from offer routing. Google: Merchant Center data, and UCP only for real agentic-commerce goals. https://developers.google.com/merchant/ucp

Myth checks

Treat as unsupported or overgeneralized unless fresh evidence proves otherwise:

• "AEO is mostly PR." / "AEO is mostly owned content." / "Reddit is always the answer."
• "Track a few prompts many times per day and you have reliable AEO measurement."
• "Optimize for the exact user prompt wording."
• "One Google AI surface tells you how all Google AI surfaces behave."
• "A citation is the same as brand visibility."
• "Direct AI referral clicks capture AEO value."
• "One shopping carousel is a rank."
• "llms.txt is required for Google AI visibility." / "There is a special AEO schema that guarantees citations."
• "Every page should be broken into tiny answer chunks." / "More FAQ pages automatically increase AI visibility."
• "Classic SEO no longer matters." / "Ranking in classic search guarantees AI citation."

Prioritization

Score proposed actions 1-5 on expected impact, evidence confidence, reach across priority prompt clusters, durability, implementation effort (reverse-scored), and dependency risk (reverse-scored).

Prefer durable, high-confidence bottleneck fixes over speculative rewrites.

Default recommendation buckets: fix access/technical eligibility; strengthen or consolidate existing content; create genuinely new information value; improve entity/product/local data consistency; earn presence on already-relevant third-party sources; improve freshness/discovery/distribution; improve measurement/experimentation; add commerce/agent readiness only when economically relevant.

Output contracts

Lead with the answer or decision. Separate observed evidence, inference, recommendation, and unknowns.

Use this structure for a full audit, and adapt it for smaller deliverables:

AEO Audit — [Brand / Domain]

Executive summary
- Current state:
- Primary bottleneck:
- Highest-leverage move:
- Confidence:

Market baseline

| Dimension | Brand | Competitor(s) | Benchmark | Interpretation |
| --- | --- | --- | --- | --- |
| Visibility / Share of Voice |  |  |  |  |
| Mention Position |  |  |  |  |
| Citation Share |  |  |  |  |
| Brand/company source share |  |  |  |  |
| Earned/institution share |  |  |  |  |
| Social/UGC share |  |  |  |  |

Visibility-stack diagnosis

| Layer | Evidence | Status | Why it matters |
| --- | --- | --- | --- |
| Access/eligibility |  | Pass/Watch/Fail |  |
| Retrieval/fan-out fit |  |  |  |
| Citation readiness |  |  |  |
| Brand visibility |  |  |  |
| Narrative/accuracy |  |  |  |
| Business outcome |  |  |  |

Priority actions

| Priority | Action | Bottleneck | Impact | Confidence | Effort | Dependency | Owner |
| --- | --- | --- | --- | --- | --- | --- | --- |
|  |  |  |  |  |  |  |  |

Measurement plan
[Prompt cohort, engines/models, languages/markets, cadence, controls, metrics]

Evidence & caveats
[Separate official guidance, observational research, site evidence, assumptions]

Other deliverables must contain at least these fields:

• Prompt portfolio: demand model (cluster, user job, stage, demand evidence, business value, priority); portfolio (prompt family, research-opening prompt, variants, fact/entity needs, likely fan-outs, engines, market/language); measurement groups (stable core, discovery cohort, diagnostic controls, repeated-run experiments).
• Citation gap: current source market by engine/language/market; per-cluster loss table (mentioned, cited, winners, source category, fan-out need, gap type); citation-neighbor map (domain, co-citation share, role, reachability, action); interventions across owned, evidence, ecosystem, and entity/technical/freshness; before/after test design.
• Answer accuracy & framing: claim ledger (engine/prompt, claim, source, correct?, fresh?, solicited or editorial, decision impact, severity); source-of-truth gaps (fact, canonical source, problem, required update, distribution path); framing summary of positive/negative frames, missing differentiators, vague owned claims, and borrowed competitor frames.
• Content brief: target user job; prompt cluster with fan-out needs and target engines/markets; current answer/source landscape; information-gain requirement; required evidence; recommended structure; narrative risk checks; success criteria.
• Roadmap: strategy thesis; opportunity map (theme, demand, visibility, citation gap, narrative gap, business value, priority); workstreams; experiments with hypothesis, target segment, bottleneck, intervention, control, sample/cadence, metric, and stop/scale condition.
• Practice operating system: mission; loop table (loop, owner, cadence, inputs, trigger, output, KPI); decision rules.

Final-answer contract

Lead with the diagnosis or decision, not a giant checklist.

Every substantive deliverable should include, as applicable: executive finding; evidence with source/date context; diagnosed bottleneck(s); engine/model/market segmentation; prioritized actions with expected impact, confidence, effort, dependencies, and owner when useful; measurement/test design; and caveats distinguishing official guidance, observational evidence, inference, and unknowns.

Quality gate

Before finalizing, verify that the work:

• identifies a bottleneck rather than reciting generic best practices
• names the target engine(s)/surface(s)
• separates official guidance from observational findings
• cites/dates changing claims when research is used
• distinguishes citations from visible brand mentions
• benchmarks source mix before prescribing channels
• treats prompt-portfolio design and model volatility as measurement issues
• separates visibility, citation influence, narrative/accuracy, and business impact
• includes a measurement plan
• avoids unsupported AEO myths and deterministic ranking claims
• ends with prioritized next actions