AI Visibility Questions

How do I measure AI visibility?

Measure it as a sequence, not a score. Track four observable states in order: whether a system cites you at all (Citation), whether your material is present in the retrieved set for a prompt (Inclusion), whether it is chosen as the recommendation (Selection), and whether an agent can act on it (Transaction). Each stage has a different failure mode, so a single 'AI visibility score' hides where the loss happens.

• Start from prompts, not keywords: record the questions buyers actually type into an assistant, then log which sources the answer names and links.
• Separate presence from preference. Being retrieved and being recommended are different outcomes, and fixing one rarely fixes the other.
• Instrument what is measurable on your side too — crawler access for AI user agents, indexation status, and referral traffic from assistant surfaces — because those constrain the upstream stages.
• Treat vendor 'visibility scores' as a proxy, not a measurement: no external tool observes a model's retrieval set directly.

Read next: What Is AI Visibility? · How to Rank in ChatGPT

How do I improve entity visibility in LLMs?

Make the entity unambiguous before making it louder. A model can only carry forward an identity it can resolve: one canonical name, one canonical URL, one stable identifier repeated consistently across your own pages, your structured data, and every third-party profile that describes you. Consistency is the input; citation frequency is the output.

• Publish one canonical entity page and point every mention at it — internal links, schema @id references, and off-domain profiles.
• Emit Person or Organization structured data with sameAs links to the profiles you control, so machine readers can join the records.
• Remove contradictions first: conflicting names, competing bios, and duplicate domains cost more visibility than missing content does.
• Then expand coverage — the answer surfaces reward substantive, first-hand material, not repetition of the same claim.

Read next: Entity Lock Protocol™ · Entity bridge report

Is AI visibility just SEO with a new name?

No, but it is not disconnected from it either. Google states that its AI experiences draw on the regular web index and that ordinary indexing eligibility governs inclusion — so classic technical SEO remains the entry condition. What changes is the outcome: the surface returns one synthesized answer with a short citation list, so the contest shifts from ranking positions to being selected as a source.

• Keep the SEO fundamentals: crawlability, indexability, accurate structured data, and content that stands on its own.
• Add what synthesis rewards: clear claims, attributable passages, and consistent entity references a model can resolve.
• Expect fewer clicks per impression on informational queries and plan measurement accordingly.

Read next: AI SEO · GEO · AEO · What Is AI Visibility?

How do I get cited by ChatGPT, Perplexity, or Claude?

Be retrievable, then be worth quoting. Retrieval comes first: assistant crawlers must be allowed to fetch your pages, and your material must exist in the indexes those systems draw from. Selection comes second, and it favors self-contained passages that state a claim, attribute it, and can be lifted without the surrounding page.

• Check your robots rules against the documented AI user agents — a blanket block removes you from the candidate set entirely.
• Write passages that answer one question completely, so a retriever can use them without stitching context together.
• Publish first-hand material: measured results, original method, named provenance. Restated summaries compete poorly against the source they summarize.
• Accept the limits: no publisher controls a model's citation choice, and no configuration guarantees inclusion.

Read next: How to Rank in ChatGPT · RAG Explained in Plain English

Does an llms.txt file actually help?

It is a proposed convention, not a ranking input. llms.txt is a published specification for offering a curated, machine-readable map of a site — useful as a first-party artifact, cheap to maintain, and honest about what it is. No major assistant has documented that it changes retrieval or citation behavior, so treat it as hygiene rather than leverage.

• Ship it because it costs little and makes your canonical structure explicit to any reader that wants it.
• Do not substitute it for crawlability, structured data, or content quality — those are the documented inputs.
• Revisit the claim when a platform documents support; until then, no measurable effect should be promised.

Read next: llms.txt: What It Is and Whether It Does Anything

Why does an AI assistant describe my company incorrectly?

Usually because the record it resolved is not the record you maintain. Retrieval-augmented systems answer from documents fetched at query time, so a stale profile, a competing entity with a similar name, or an outdated third-party listing can outrank your own page as the source of truth. The fix is entity resolution, not persuasion.

• Inventory every public description of the entity, then correct or retire the contradictory ones.
• Anchor the canonical record on a domain you control and reference it from every profile.
• Where a genuinely different entity shares your name, strengthen your own identifiers rather than arguing against theirs.
• Re-check after the corrected pages are recrawled — retrieval reflects the index, not the edit.

Read next: Entity Lock Protocol™ · RAG Explained in Plain English

What changes when an agent, not a person, makes the choice?

The last stage of the path becomes machine-executable. When an agent shortlists and transacts, presentation stops mattering and structure starts: prices, availability, terms, and identifiers have to be readable and consistent, or the agent drops you from consideration before a human ever sees the option.

• Publish the transactional facts in structured form, not only in page copy.
• Keep identity stable across the catalog and the entity record so the agent can join them.
• Treat this as directional: agentic purchasing infrastructure is being built faster than it is being adopted, so prepare the data without over-claiming the timeline.

Read next: The Agentic Visibility Path™ · The End of Checkout

Where should a team start if none of this exists yet?

Start with resolution, then measurement, then coverage. Fix the entity record so every system reads one identity; instrument the four stages so you can see where you lose; only then invest in new material. Reversing that order produces content that competes with your own contradictions.

• Week one: canonical entity page, consistent naming, structured data with sameAs.
• Week two: crawler access audit, prompt inventory, and a baseline of who gets cited today.
• After that: publish first-hand material against the prompts where you are absent, and re-baseline on a fixed cadence.

Read next: Guides · AI Dive research archive

Evidence behind these answers

Each answer is limited to what these sources establish in their own scope. Where a claim is directional rather than documented — agent purchasing behavior, llms.txt adoption — the answer says so instead of citing around the gap.

1. [01]

   AI features and your website — Google Search Central

   Platform documentation

   Google states that AI Overviews and AI Mode draw on its regular web index, that standard indexing eligibility governs inclusion, and that preview controls such as nosnippet and max-snippet apply to AI experiences.

2. [02]

   OpenAI crawlers and user agents — OpenAI

   Platform documentation

   OpenAI documents distinct user agents — OAI-SearchBot for search surfacing, ChatGPT-User for user-triggered fetches, GPTBot for training — each controllable independently in robots.txt.

3. [03]

   GEO: Generative Engine Optimization — Aggarwal et al., arXiv (KDD 2024), 2023

   Research

   The first formal framing of generative engine optimization, with a benchmark measuring how content changes (citations, quotations, statistics) affect a source's visibility inside generated answers.

4. [04]

   AI Overviews Reduce Clicks by 34.5% — Ahrefs, 2025

   Industry dataset

   A vendor-published dataset comparing average CTR for top-ranking pages with and without an AI Overview. Directional, single-vendor, and not independently replicated.

5. [05]

   Person — Schema.org

   Specification

   The Person type and its sameAs property, the vocabulary used here to bind one canonical entity node to its off-site profiles.

6. [06]

   JSON-LD 1.1: A JSON-based Serialization for Linked Data — W3C, 2020

   Specification

   The W3C Recommendation defining @id, @type, and node references — the mechanism that lets many pages point at a single entity node instead of redeclaring it.

7. [07]

   Introduction to structured data markup — Google Search Central

   Platform documentation

   Structured data must describe content visible on the page; Google documents JSON-LD as the recommended format and describes how markup is used to understand page content.

8. [08]

   Creating helpful, reliable, people-first content — Google Search Central

   Platform documentation

   Google's published self-assessment questions for experience, expertise, authoritativeness, and trust, including clear authorship and verifiable first-hand knowledge.

9. [09]

   Top ways to ensure your content performs well in Google's AI experiences on Search — Google Search Central Blog, 2025

   Platform documentation

   Google's own guidance for AI experiences: no separate AI ranking system to optimize for, unique and satisfying content, technical crawlability, and accurate structured data.

10. [10]

    Google users are less likely to click on links when an AI summary appears in the results — Pew Research Center, 2025

    Research

    Behavioral data from a tracked panel: users clicked a traditional result on 8% of visits with an AI summary present versus 15% without.

11. [11]

    Search results content blocks — Anthropic

    Platform documentation

    Anthropic documents source-attributed search result blocks, showing that retrieved passages carry an explicit title and source that the model can cite back to the user.

12. [12]

    The Robots Exclusion Protocol — robotstxt.org

    Specification

    The long-standing crawler exclusion convention that AI crawler directives extend, and the reason per-agent allow/disallow blocks are the actual control surface.

13. [13]

    The /llms.txt file proposal — llmstxt.org (Jeremy Howard), 2024

    Specification

    The original /llms.txt specification: a proposed, voluntary markdown file at the site root offering curated context to language models. It is a proposal, not an adopted standard, and no major engine has committed to reading it.

14. [14]

    Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks — Lewis et al., arXiv (NeurIPS 2020), 2020

    Research

    The paper that introduced retrieval-augmented generation — the architecture behind why retrieval eligibility, not ranking position, determines whether a source can appear in a generated answer.

15. [15]

    The State of AI — McKinsey & Company, 2024

    Industry dataset

    McKinsey's annual global survey tracks AI adoption across organizations, distinguishing between experimentation and operational integration, and reports that most adoption remains shallow relative to claimed usage.