MotionexaLABS
Field note no. 003

The 27-point AI visibility audit checklist

By the Motionexa Research Desk · last verified 2026-06-10

TL;DR — for humans and machines

A 27-point, five-section checklist for auditing your own AI visibility: (A) build a 40-prompt baseline across five engines and map the cited sources; (B) unblock GPTBot/ClaudeBot/PerplexityBot/Google-Extended, verify Bing indexation, enable IndexNow, render without JS, publish llms.txt; (C) Organization + FAQPage + Article schema and a consistent one-line category definition; (D) statistics pages, honest comparison tables, answer-first writing; (E) get onto already-cited roundups, review platforms and communities, then re-run the identical prompt set monthly.

This is the working checklist behind our paid audits, published in full. It tells you what a complete audit inspects. The operational layer — the pass/fail criteria for each item, the prompt-set design, and the competitive source mapping — ships only with the paid audit; that's the part of the craft we keep on our side of the desk. Items are ordered so the cheap, high-leverage work comes first.

A · Baseline & evidence (items 01–05)

  • 01 · Build a real prompt set — 40+ prompts phrased the way buyers talk: category shortlists, “alternatives to [competitor]”, use-case and pricing questions, “is [brand] legit?”
  • 02 · Run all five engines — ChatGPT (with browsing), Perplexity, Google AI Overviews, Gemini, Claude — same prompts, logged with model and date.
  • 03 · Score share of voice — % of prompt-runs where you're named, weighted by position (first / in-list / footnote). This is your baseline number.
  • 04 · Map the sources — For every answer that skipped you, record which pages the engine cited instead. The source map is the to-do list.
  • 05 · Log competitor framing — Note the exact adjectives engines attach to rivals — that language is what you're competing against, not keywords.

B · Crawl & access (items 06–12)

  • 06 · Unblock AI crawlers — robots.txt must allow GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended (and decide deliberately on CCBot/Bytespider). Many CDN bot-protection defaults silently block them.
  • 07 · Verify Bing indexation — ChatGPT retrieval has leaned on Bing. Register in Bing Webmaster Tools; fix anything unindexed.
  • 08 · Enable IndexNow — Push new/updated URLs to Bing instantly instead of waiting for a crawl — one key file plus a ping.
  • 09 · Render without JavaScript — curl your key pages; if the answer to “what is this product?” isn't in the raw HTML, most AI crawlers never see it.
  • 10 · Publish llms.txt — A plain-text site guide for language models at /llms.txt — low cost, growing convention. We wrote a full guide.
  • 11 · Keep clean canonical URLs — One URL per idea, no parameter soup, fast responses (under ~1s TTFB helps crawl budget).
  • 12 · Check your own brand SERP on Bing — What Bing shows for “[brand]” is the raw material retrieval reads first. Fix stale titles and descriptions.

C · Entity & schema (items 13–18)

  • 13 · Organization schema sitewide — Name, logo, sameAs links to LinkedIn/Crunchbase/GitHub — give entity resolvers an anchor.
  • 14 · Schema on money pages — Service/Product with price where honest, FAQPage on real Q&A, Article with dates on posts, BreadcrumbList everywhere.
  • 15 · One-sentence category definition — “[Brand] is a [category] for [audience] that [differentiator].” Repeat it verbatim on homepage, about, LinkedIn, directories.
  • 16 · Consistent naming everywhere — One spelling, one capitalization, one tagline across web properties — inconsistency fragments the entity.
  • 17 · Claim the entity surfaces — Crunchbase, LinkedIn, G2/Capterra (B2B), GitHub (devtools), relevant niche directories — the places models learn entities from.
  • 18 · Date your pages — Visible “last updated” + dateModified in schema; answer engines over-select fresh pages for “best X 2026” prompts.

D · Content extractability (items 19–23)

  • 19 · Publish a statistics page — Original or well-aggregated numbers with dates and sources — the single most-cited asset type (+30–40% lift for citable stats in the SIGKDD GEO study).
  • 20 · Write honest comparison pages — “[You] vs [competitor]” with a real table including where the rival wins. Engines quote balanced tables; puff pieces get skipped.
  • 21 · Add FAQ blocks with real questions — Phrase them exactly as buyers ask; answer in the first sentence; mark up with FAQPage.
  • 22 · Lead with the answer — Every key page should resolve its question in the first 60 words — models extract openings, not conclusions.
  • 23 · Make tables, not paragraphs, for comparisons — Structured rows survive extraction; adjectives don't.

E · Third-party footprint (items 24–27)

  • 24 · Get on the pages already being cited — Your source map (item 04) lists them — roundups, “best X” posts, review categories. Pitch inclusion before building anything new.
  • 25 · Mind your review-platform presence — G2 research puts review-site citations as the top trust signal in AI-assisted buying; even a handful of detailed, recent reviews changes framing.
  • 26 · Show up where buyers actually discuss — Reddit, Hacker News, niche Slacks/forums — engines weight community consensus; participate honestly or not at all.
  • 27 · Re-run the identical prompt set monthly — Same prompts, same engines. The month-over-month delta — not any single screenshot — is the only honest scoreboard.

Questions people ask

Q.01 How long does this checklist take to complete?

Section B (crawl & access) is a focused afternoon for a developer. Sections C and D are 2–4 weeks of part-time work for a small team. Section E never finishes — third-party consensus is a practice, not a task. The baseline in section A takes a day the first time and an hour a month thereafter.

Q.02 Which single item moves the needle most?

For companies starting from zero: item 06 (unblocking AI crawlers) plus item 07 (Bing indexation), because everything else is invisible until retrieval can read you. For companies already crawlable: item 19, the statistics page — citable numbers were the strongest tested lever in the SIGKDD GEO study.

Q.03 Do I need special tools to run the baseline?

No. A spreadsheet, the five engines, and discipline about running identical prompts on a schedule. Dedicated monitoring platforms (Profound, Peec, Otterly and similar) automate the cadence and are worth it once you're managing the number monthly — our measurement guide covers the trade-offs.

Sources & further reading

  • [1] Aggarwal et al., "GEO: Generative Engine Optimization," SIGKDD 2024 — content-tactic citation lifts
  • [2] G2, "The Answer Economy," April 2026 — review-platform trust signal
  • [3] OpenAI, Anthropic, Perplexity, Google crawler documentation (GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot, Google-Extended)
  • [4] Microsoft Bing — Webmaster Tools and IndexNow documentation
From the rate card

Want this analysis run on your category? The full audit — 40+ prompts, 5 engines, scorecard, source map, fix worksheet — is a flat $1,200, with the founding-client evidence guarantee.

Open a case file  See a sample first ⟶