A wallchart for AI visibility · Edition 2026.2 · audited 2026-10-02
Periodic Table of AI Visibility
Sixty-nine things that decide whether an answer engine names you. Grouped by what they do: the surfaces that cite, the plumbing that lets them read you, the signals they weigh, the artifacts worth citing, the channels that carry them, the numbers that tell you it worked, and ten inhibitors to remove before you add anything.
AI visibility is whether an answer engine (ChatGPT, Perplexity, Claude, Gemini, Copilot, Grok, Google AI Overviews and AI Mode) names your company when a buyer asks a category question. It covers four outcomes: ranking (your position in a list), citation (a linked source inside the answer), mention (named, no link) and recognition (the model knows you without looking anything up). Being listed is ranking. Being chosen is the other three.
An element is one thing you can fix on its own. A blank square is an element you do not have yet. Find your blank squares and you can predict what an answer engine does with your site before you ever ask it. Everything on this page is free and ungated, because gating the one page about being cited by AI would make it the one page AI cannot cite.
Click any element for what it is, the move, and the watch-out.
Top left of a tile is the element number. Top right is time to effect: d days, w weeks, q quarters. Outlined tiles are inhibitors: subtract, don't add. Filter by name or hide a group with the legend.
If the chart does not load here, open it full screen. Every element is also written out below, so nothing on this page depends on the frame.
Engines
the surfaces that do the citing
- CgChatGPT — The highest-volume answer surface. It names a handful of sources per answer and almost never shows the runners-up.w
- PpPerplexity — Citation-first by design — it shows its sources inline, which makes it the cheapest place on earth to check whether you are retrievable.d
- ClClaude — Answers from fewer, longer sources. Rewards a page that holds a full argument over one that covers a keyword.w
- AoAI Overviews — Google's generated answer above the ten blue links. It absorbs the click that used to go to position one.w
- AmAI Mode — Google's conversational surface. It fans one typed question into many hidden ones and assembles the answer from all of them.w
- GeGemini — Google's assistant, and increasingly the default inside Workspace, where a lot of B2B research quietly happens.w
- CpCopilot — Bing-grounded and embedded in Microsoft 365, which means it sits inside the tools enterprise buyers already have open.w
- GkGrok — Grounded heavily in live posts, so it reflects what is being said about you right now rather than what your site claims.w
Retrieval
whether a machine can read you at all
- Rbrobots.txt — The gate every crawler checks first. One over-broad disallow line can shut out the exact agents that feed the answer engines.d
- SmSitemap — The index of what exists and when it last changed. Without it, discovery depends on links you may not have.d
- Ltllms.txt — A plain-text map of what matters on your site, written for models instead of browsers. Emerging convention, cheap to ship.d
- Axagents.txt — Declares what an agent can do on your site — not merely what it may read. The capability layer above robots.txt.d
- JlJSON-LD — Structured data that states the facts of a page in a form no parser has to guess at: product, price, author, organization.w
- EtEntity — The link between your name and a thing the model already knows about. sameAs references, consistent naming, corroborating profiles.q
- CkChunk — Retrieval quotes passages, not pages. A paragraph that only makes sense after reading the three above it will not be quoted.w
- SrServer render — Content present in the HTML response rather than assembled later by JavaScript. Many crawlers never run the script.w
- FrFreshness — A visible, accurate date. Models weigh recency heavily on anything that looks like it could be stale.w
- CnCanonical — The single address a piece of content lives at. Split versions divide the signal between them and neither wins.w
- TrTranscript — A video is citable only to the extent its words exist somewhere as text. The file itself is opaque.w
- AtAlt text — The only description of an image most retrieval pipelines will ever have. Chart, screenshot, diagram — all invisible without it.d
- HdHeading map — The outline a parser builds from your H2s and H3s. It decides where one idea stops and the next begins.d
- LnInternal links — How crawlers find the page and how models infer what it is about from the words pointing at it.w
Signals
what the model weighs once it can read you
- CyClarity — Whether the first line says what the thing is in words a stranger already knows.d
- SySpecificity — Named nouns beat adjectives. A sentence with a number, a unit, and a named thing in it is quotable; one without is not.d
- PfProof — A claim with a source, a number, and a date attached. Models reproduce claims they can attribute.w
- BoBuyer outcome — The changed state the buyer ends up in, stated in their language rather than your feature list.w
- CmComparison — An honest account of how you differ from the named alternatives. Buyers ask engines to compare constantly.w
- PcPricing clarity — A number, a unit, and what changes it. 'Contact us' removes you from every price-shaped question.w
- RmRich media — Whether your video and images are citable or opaque. A player an agent cannot read scores as absent.w
- FqFAQ block — Question-shaped headings with self-contained answers — the exact shape retrieval wants to lift.d
- DfDefinition — A crisp 'X is Y' sentence near the top. Definitional queries are the highest-volume door into any category.w
Artifacts
the shapes that actually get quoted
- TdTeardown — A public, specific critique of a real example. Reliably the most-shared artifact in a technical niche.w
- BnBenchmark — A number nobody else has because nobody else measured it. The single most citable object you can publish.q
- DsDataset — The raw rows behind a benchmark. It gets referenced long after the post that announced it is forgotten.q
- CsCase study — One customer, one measured before and after. The proof artifact buyers and models both look for by name.w
- RkRunbook — Ordered steps someone can follow without you. High utility, high save rate, endlessly quotable.w
- CaCalculator — An interactive tool that returns a personal number. It earns links from people who will never read your posts.q
- ShScreenshot — Visual evidence the thing exists and works. The cheapest media that materially raises trust.d
- VoVideo — Highest-trust format on every channel, and the only one that carries tone. Worthless to retrieval without its text.w
- NiNewsletter — The one audience nobody can take away — no ranking, no algorithm, no platform between you and the reader.q
- TpTemplate — A file someone can open and use today. It converts a reader into a user before any purchase exists.w
Channels
where the artifact goes to be found
- LiLinkedIn — Graph-gated: reach comes from who already follows you, not from what people search. Documents outperform text.w
- YtYouTube — Query-gated, so it compounds instead of decaying — and it is consistently among the most-cited domains in AI answers.q
- ThThreads — Fast-moving, conversation-shaped, low production cost. Good for testing a claim before you build around it.w
- ReReddit — Disproportionately cited by answer engines because the discussion is real. Also the fastest place to get punished for marketing.q
- GiGitHub — A trusted domain where a README is a document engines read closely. Underused by non-developer companies.q
- MkMarketplace — A cloud marketplace listing — an authoritative third-party page describing your product that you mostly control.q
- PoPodcast — Borrowed audience plus a transcript on someone else's trusted domain. Two assets from one hour.q
- DyDirectory — Review and category sites. Engines lean on them heavily for comparison and shortlist questions.w
- IgInstagram — Where the person behind the company becomes legible. Weak for retrieval, strong for recognition.w
- GpGuest post — Your argument on a domain with authority you have not earned yet. Slow, but it moves the entity signal.q
Measurement
how you know any of it moved
- SvShare of voice — Of all answers to your category's questions, the share that names you. The closest thing to a scoreboard.q
- MtMentions — How often you are named at all, cited or not. Being named without a link still shapes the buyer.w
- CiCitations — Answers that link you as a source. The strongest evidence a specific page earned its place.w
- PsPrompt set — The fixed list of questions you re-ask on a schedule. Without it, every other number here drifts.d
- SbSwitchback — Turning a change on and off in alternating time windows. The only clean test when you cannot split traffic by visitor.q
- ArAI referrals — Sessions arriving from an answer engine. Small in volume, and consistently the highest-intent traffic on the site.w
- BsBrand search — People searching your name directly. It rises when awareness works, whatever the attribution says.q
- ImImpressions — How many times you were shown. Necessary context, and worthless as a goal on its own.w
Inhibitors
subtract, don't add
- BcBlocked crawler — A disallow rule aimed at scrapers that also shuts out the agents feeding the engines you want to appear in.d
- JoJS-only — Content that only exists after the script runs. Many retrieval pipelines never get that far.w
- GaGated — Your best evidence sitting behind a form. Invisible to every engine and most buyers.w
- PlPaywall — Same failure as a gate, with a card attached. The content may be excellent and it cannot be cited.w
- JwJargon wall — Language only an insider parses. It fails both the model matching intent and the buyer scanning for fit.d
- VcVague claim — 'Industry-leading', 'seamless', 'enterprise-grade'. Unfalsifiable, so unquotable — nothing to attribute.d
- DuDuplicate — The same content at several addresses, splitting whatever authority any single version might have had.w
- OpOrphan page — A page nothing links to. It exists, and for discovery purposes it does not.w
- SlSlop — High-volume thin content. It buries your own good pages and trains readers that your name means noise.q
- NdNo date — Undated content reads as possibly stale, and stale gets skipped on anything time-sensitive.d
New editions and the checklist version go to the list first.
One email a week. Unsubscribe in one click.
The questions as people type them, each answered in two sentences.
Why doesn't ChatGPT mention my company?Engines
Your buyer asked, got three names, and you weren't one of them. Each engine chooses differently, and you only know you're missing from the ones you checked.
Google's AI Overview shows my competitors, not me.Engines
The answer box took the click that used to be yours. Position one doesn't matter if the box above it names someone else.
How do I show up in Perplexity or Grok?Engines
Different engines read different things: Perplexity reads the open web, Grok reads live posts, Gemini reads inside Workspace. One fix doesn't cover all eight.
Is my website blocked from ChatGPT?Retrieval
One line in robots.txt meant for scrapers can block the crawler that feeds the answer, and your analytics will never tell you.
How do I know if AI crawlers can even read my site?Retrieval
If there's no sitemap entry and nothing links to a page, no engine finds it. Existing is not the same as being found.
What is llms.txt and do I need it?Retrieval
It's a note taped to the front door for robots. Cheap to make; Google says it ignores it. Worth an hour, not a week.
Why does AI quote my competitor's page and not mine?Retrieval
Same claim, same product, theirs gets picked. Usually the page the engine could actually read: rendered on the server, question-shaped headings, each section standing on its own.
ChatGPT gives outdated information about my product.Retrieval
The old page won. Three copies with no canonical, or no honest updated date, and the engine picks the stale one.
My videos and screenshots never show up in AI answers.Retrieval
Because there are no words in them. A video with no transcript page and a chart whose alt text says 'chart.png' are blank to retrieval.
ChatGPT describes my product wrong.Signals
Your first line says 'we empower modern teams.' The model had to guess what you do, and it guessed from someone else's page.
AI won't repeat my claims or my numbers.Signals
'Fast onboarding' can't be quoted. 'Live in 4 days for teams under 50' can. No number, no source, no citation.
When someone asks AI to compare me to a competitor, I lose.Signals
Because you never wrote the comparison. The review site with ads did, and the engine cited them.
AI lists my competitors with prices and leaves me out.Signals
'Contact us for pricing' removed you from every 'how much does it cost' answer. A starting price puts you back in.
AI answers questions about my product from a review site's FAQ instead of mine.Signals
Your support tickets already hold the real questions. Ten of them, answered in two sentences each, get lifted word for word.
When someone asks 'what is [my category]', AI defines it using a competitor.Signals
The definition line at the top of a page is the front door. If yours isn't there, theirs is.
What kind of content actually gets cited by AI?Artifacts
Evidence, not opinion. A teardown, a benchmark, a dataset, a measured case study, a runbook. Most B2B sites have none.
Why don't my blog posts show up in AI answers?Artifacts
Fifty posts of opinion are fifty pages with nothing attributable. One benchmark with a method and a date beats all of them.
AI cites a competitor's study in my category.Artifacts
They published the number. You can publish yours: fifty samples, a stated method, a date. Small is fine.
Do videos help with AI visibility?Artifacts
Only with words attached. The video is the trust layer; the transcript is the retrieval layer. You need both.
AI never links to my tools or calculators.Artifacts
Because they're gated. The moment a calculator asks for an email, it stops being linked. Ungate the tool, gate the custom result.
Does a newsletter or a template help me get cited?Artifacts
They build the audience that later types your name into an engine. Recognition first, citation later.
Does posting on LinkedIn help AI visibility?Channels
Almost nothing posted there gets cited. It's where you're seen, not where you're cited. Different job.
Why does AI cite Reddit over my own website?Channels
Because the discussion is real. One honest thread can outrank your whole site, and one sales pitch gets you banned.
Does YouTube or GitHub get cited by ChatGPT?Channels
YouTube is consistently among the most-cited domains, and a README on GitHub is a document on a trusted domain. Both read closely.
Do G2 or Capterra reviews affect AI answers?Channels
'Best X tools' answers lean on review sites. An empty profile reads as an inactive vendor.
My AWS Marketplace listing doesn't show up in AI answers.Channels
Treated as a form it tells engines nothing. Treated as a landing page, with a definition, proof and a price, it gets cited for 'where can I buy X'.
Do guest posts and podcasts help with AI visibility?Channels
A transcript page on someone else's trusted domain is a citation you borrowed. No transcript, no retrieval value.
How do I track AI visibility?Measurement
Thirty real buyer questions, frozen, asked monthly. Named in 6 of 30 is 20% share. That's the scoreboard.
How do I measure ChatGPT mentions of my brand?Measurement
Mentions come before citations. Track named-but-unlinked separately; it's the first sign something started working.
AI search traffic isn't showing in my analytics.Measurement
Many AI visits arrive as 'direct.' Segment the ones you can see; they're under 2% of traffic and the highest-intent cohort on the site.
My site disappeared from ChatGPT.Inhibitors
A 'block all bots' rule meant for scrapers blocked the crawlers that feed the answer. Months invisible, five-minute fix.
My pricing page shows up blank to AI.Inhibitors
It's built in JavaScript. To most crawlers the page is a header and nothing else. View source and look for your main claim.
Does gated content hurt AI visibility?Inhibitors
Your best proof behind an email form is invisible to every engine and most buyers. Ungate the proof; gate the tailored thing.
Does duplicate content hurt AI search?Inhibitors
The same article at three addresses splits authority three ways. None of them wins.
Does AI-generated content hurt AI visibility?Inhibitors
Fifty thin posts buried the three good ones and taught the engines your name means noise. Deleting is faster than waiting.
Do pages without dates get cited?Inhibitors
Undated reads as possibly stale, and stale gets skipped on anything time-sensitive. Old but dated beats new but undated.
SEO, AEO, GEO, LLMO: four outcomes, one word.
| Term | Means | Bought by |
|---|---|---|
| Ranking | your position in a list | SEO |
| Citation | a linked source inside the answer | AEO · GEO (they overlap almost completely) |
| Mention | named, no link — usually the first sign something is working | the same work, earlier |
| Recognition | the model knows you without looking anything up; comes from training data, so it is slow | LLMO · AIO |
| AI visibility | all four. Listed is ranking. Chosen is the other three. | the whole table |
"AI SEO" means either SEO for AI engines or using AI to do SEO. Ask which before you agree with anyone.
The table as eight installable AI skills for Claude, ChatGPT, Gemini or Copilot.
Each skill is a self-contained prompt that owns one row of the table and routes to the next. They use only what you paste; none of them invents a number. Install guides for every model are in the repository, and the whole set is one paste-in file.
The Blank-Square Audit
Scores a page against all 69 elements and names the one skill to run next. Run it first.
Inhibitor Sweep
The ten things to remove, with the exact check for each.
Retrieval Gate Check
Fourteen pass/fail gates a page must clear before any engine can read it.
Citable Sentence Rewriter
First line, strongest claim, pricing line, headings, definition, real-ticket FAQ.
Machine-Readable Files
llms.txt, agents.txt and JSON-LD from facts you supply.
Artifact Picker
Which of the ten citable artifacts to make next, given an afternoon, a week or a quarter.
Channel Router
Seen versus cited: where each artifact goes and in which form.
Prompt Set and Scoreboard
Thirty frozen buyer questions and the monthly numbers with unit and direction.
The AI Visibility Audit Kit: the scorecard, the tracker and the files, ready to open.
The table tells you what the 69 squares are. The kit is how you score your own, in order, and measure whether anything moved. Files you open and use the same day, not a course.
- The 69-element scorecard workbook. Mark each square Present, Blank or Inhibitor present. It ranks your open squares by time to effect, days before weeks before quarters, inhibitors doubled.
- The prompt-set tracker. Thirty frozen buyer questions, a monthly run log, and a scoreboard that computes share of voice, mentions, citations, AI-referral conversion and a switchback test, with the unit and direction for every number.
- Fill-in templates for llms.txt, agents.txt, the robots.txt allow-list, and JSON-LD for Organization, Product and FAQPage.
- The weekend inhibitor runbook. Ten things to remove, with the exact command or check for each.
$99
Sent to the email address you use at checkout. The table, the data and the eight skills stay free.
Get the kitWant it run for you instead? Book a free 30-minute call.
A catalog tells you what exists. A periodic table tells you why it gets picked.
I'm Ron Davis. Eight years inside AWS Marketplace in technical business development, covering go-to-market for more than 3,000 AWS partners. The last eighteen months were spent on products where the buyer is an AI agent, not a person. Listings written for humans were unreadable to the agent. Being listed was not being chosen.
Before that: service catalogs at Accenture and CompuCom, 28,000 knowledge articles centralized and de-duplicated at a bank (that job is the Duplicate and Canonical tiles), 1,000 tasks standardized across 8 service offerings at Dell. This is the first catalog I built with a why.
The receipt behind the Signals row: in 2026 I read 595 AWS Marketplace listings the way a machine reads them. 482 of the 595 opened with a superlative, and not one superlative is quotable by a machine.
Mendeleev left blank squares for elements nobody had found yet and said what would fill them. I'm not a chemist. I'm a navigator. An element here just means one thing you can fix on its own.