AI visibility
How answer engines decide whom to name, how to measure that with intervals instead of single runs, and the practices that move it: retrievable pages, cited sources, a consistent entity.
21 lessons in 4 modules · for brand and growth teams asked “are we in chatgpt?” · each lesson shows its evidence tier or its basis
Academy AI visibility
Open a lesson to read what to do, why, and how the product measures it.
Module 1 of 4
Measure it honestly
Sampling, intervals, sample size and significance, with the numbers computed live.
What to do: Run each tracked prompt several times per engine and report visibility with a confidence interval and sample size.
Why (practitioner): LLM answers are stochastic. A single run per prompt cannot tell a change from noise.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Wilson intervals per prompt, engine and week; two-proportion tests for changes.
AI visibilityWhat to do: Never report a rate without its interval and n. Named in 3 of 5 runs is 60%, 95% interval 23%–88%, n = 5.
Why: Answer engines are sampled, not read: ask the same prompt twice and you get two answers. A single run is one draw from a distribution, and the interval says how wide that distribution could be.Statistics: Wilson score interval
How AwesomeSEO measures it: Every rate in AI visibility is a Wilson interval from @awesomeseo/evidence, shown as point, range and n.
AI visibilityWhat to do: Decide the precision you need before you measure. ±10 points at 95% confidence needs 93 runs per prompt; ±5 needs 381. Pool prompts by intent when a single prompt cannot afford that.
Why: Precision grows with the square root of n: four times the runs halves the interval. Planning the sample stops a team from acting on a difference the data cannot show.Statistics: Wilson half-width, worst case p = 0.5
How AwesomeSEO measures it: The sample-size tool computes this for any precision and expected rate; the product samples five runs per prompt per engine per week and pools across weeks.
Sample-size calculatorWhat to do: A move from 51% to 63% on 35 runs a week is not significant (p = 0.33) and is shown as n.s. The same difference held over eight weeks is significant (p = 0.01).
Why: Weekly wobble is the normal state of a sampled rate. Reporting direction without a test turns noise into decisions.Statistics: two-proportion z-test
How AwesomeSEO measures it: Every delta in the product carries a two-proportion test; counts (citations, bot visits) carry an exact count-change test.
AI visibilityModule 2 of 4
Be retrievable
Answer engines fetch pages at request time. Decide who may read what, and keep it readable.
What to do: Set robots.txt rules per AI user agent and path group on purpose: training crawlers, search-index crawlers and live user fetches are separate tokens (for example GPTBot, OAI-SearchBot and ChatGPT-User).
Why (practitioner): OpenAI, Anthropic and Perplexity document separate user agents; blocking a search-index bot can keep your pages out of that engine’s answers, while blocking a training bot does not. It is a business decision, so make it explicitly.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Access matrix per bot and path group from robots.txt (RFC 9309 parser), plus AI-bot visits per URL from server or CDN logs.
Site healthWhat to do: Serve prices, stock, shipping, returns, reviews and FAQs in the server HTML, ungated and fast, with structured data that matches the page.
Why (practitioner): Answer engines increasingly fetch pages live to ground what they say. Pages that are slow, gated, JavaScript-only or unstructured get skipped or misquoted.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Raw vs rendered HTML diff, TTFB per template, schema parity and crawlable-FAQ checks.
Site healthWhat to do: Server-render or pre-render pages that should rank; check that text, links and schema are in the HTML response.
Why (confirmed): Google renders JavaScript, but later and at a cost; many AI crawlers do not render it at all.Google states it in documentation or official guidance.
Site healthWhat to do: Definitions first, claim then evidence, tables for comparisons, clear headings, entities named in full.
Why (practitioner): Low cost of retrieval for retrieval-augmented engines.A working heuristic from experienced SEOs. We test it on your own data.
ContentWhat to do: Refresh the pages that get cited, change what is actually out of date, and show a visible, truthful updated date.
Why (practitioner): Search-first engines show a date next to many citations and prefer recent pages for time-sensitive prompts in sampled answers. How much recency weighs is a practitioner model; the product tests it by comparing update dates of your cited and uncited pages. Changing only the date is excluded.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Update age of cited vs uncited pages per engine.
ContentWhat to do: Publish a short llms.txt that points to the canonical pages.
Why (speculative): Cheap to add. No major engine has confirmed using it.A hypothesis. Shown with that label, never as advice.
Site healthModule 3 of 4
Be cited
Engines lean on the sources they already trust. Get onto them.
What to do: Find the domains and pages each engine cites for your prompts and earn a presence there.
Why (practitioner): Answers are assembled from cited sources; being on them is often faster than outranking them.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Cited-domain share per engine.
AI visibilityWhat to do: Run digital PR for brand mentions with context on reputable, frequently cited sources, not just for links.
Why (practitioner): Models summarise a brand from what others say about it (reviews, mentions, citations) as well as from what it says itself.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Cited-domain share per engine and new mentions on those domains.
AI visibilityWhat to do: For queries where Google shows an AI Overview above your result, open the matching section with a direct, sourced answer and keep the page in the top results for that query.
Why (practitioner): Google states that AI Overviews draw on its core ranking systems and need no special markup. Which pages an Overview links, and how closely that follows organic rank, is a practitioner observation; the product measures it per query from a SERP sample of your own tracked queries.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Share of tracked queries with an AI Overview, the impressions they carry, and whether your domain is among the Overview’s linked sources.
Search performanceWhat to do: Map each tracked prompt to the search query it implies and work that query in Search Console like any other: position, CTR and coverage.
Why (practitioner): Search-first engines (Perplexity, Copilot, AI Overviews, AI Mode and Gemini with grounding) run a web search before they answer, so what ranks is what they can read. How strongly rank predicts a citation differs by engine and is a practitioner model; the product tests it by comparing your Search Console positions with your sampled citations.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Citation rate per prompt against the organic position of the matching query.
AI visibilityWhat to do: Verify the site in Bing Webmaster Tools, submit the sitemap, and check that money pages are indexed and rank in Bing for the queries behind your tracked prompts.
Why (practitioner): Microsoft Copilot answers with Bing search behind it, and OpenAI has said ChatGPT search draws on third-party search providers. A page missing from Bing cannot be cited by either. How Bing rank translates into citations is a practitioner model; the product tests it on your own sampled answers.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Bing index coverage and rank for tracked queries (Bing Webmaster Tools) against citation rate on Copilot and ChatGPT.
AI visibilityWhat to do: Claim and keep current the Trustpilot, Prisjakt and Google review profiles; ask real customers for reviews and reply to every one.
Why (practitioner): Review sites appear among cited domains for brand prompts in sampled answers, and engines summarise reputation from them. The weight per engine is a practitioner model; the product tests it against your sampled answers and their citations. Self-written or incentivised reviews are excluded.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Review-site share of citations per engine; review count and score per profile.
Brand & authorityWhat to do: Take part, as yourself and disclosed, in the Reddit threads, forums and review sites that already answer your prompts; answer questions, correct facts, never seed fake praise.
Why (practitioner): Forum and review domains are among the sources the engines cite most for comparison prompts in sampled answers, and OpenAI and Google both license Reddit content (2024). That more mentions there raise your mention rate is a practitioner model; the product tests it by tracking your cited-domain share per engine over time. Fake engagement is excluded.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Cited-domain share for community and review domains per engine, and your presence on the threads they cite.
Brand & authorityModule 4 of 4
Be a known entity
The same description everywhere, in Google’s own data and in the sources models learn from.
What to do: Keep the Business Profile, Merchant Center and sameAs identifiers complete and consistent with the site: name, founding year, address, category, products.
Why (practitioner): Google says AI Overviews and AI Mode are built on Search, and Gemini can ground answers with Google Search; in sampled answers Gemini and AI Overviews repeat Business Profile facts verbatim, right or wrong. That a complete profile raises mention rate is a practitioner model; the product tests it on your sampled answers and flags answers that contradict your site.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Profile completeness and consistency checks, plus contradictions between sampled answers and your site.
Brand & authorityWhat to do: One canonical description, facts and category used in About, press releases, social bios, schema and directory listings.
Why (practitioner): Language models learn a brand from repeated, consistent descriptions across sources.A working heuristic from experienced SEOs. We test it on your own data.
How AwesomeSEO measures it: Brand canon consistency score across connected profiles.
Brand & authorityWhat to do: Make sure the brand is described the same way on Wikipedia-grade references, trade press and your own About page, and keep those descriptions current.
Why (speculative): Models answer some prompts from training data rather than live retrieval, and ChatGPT and Claude do so by default for many questions. Which sources end up in a training corpus and how they are weighted is not public, so this is a hypothesis. The product measures only the outcome: mention rate in answers that cite no sources.A hypothesis. Shown with that label, never as advice.
How AwesomeSEO measures it: Mention rate in answers without citations, per engine.
AI visibilityWhat to do: Offer clean, machine-readable endpoints (feeds, APIs or an MCP server) for catalogue, prices, availability and policies.
Why (speculative): Assistants that act for shoppers need structured, current data. How engines will consume it is not settled yet.A hypothesis. Shown with that label, never as advice.
ShopifyNext track
Practise on the demo workspace: AI visibility.
Example data for a fictional Swedish brand, every number with its source, interval and n.