agentleFS
fabio-bergmannfabio-bergmann/seo-geo-skills/skills/ai-visibility-check/SKILL.md

ai-visibility-check

Skill1 starsChanged 13 days agoMITView on GitHub

Checks how AI answers treat a brand by asking a fixed set of buyer questions to the AI models available, then counting apart whether the brand is named, recommended or cited, who is named instead and which pages the answers cite. Turns cited third-party pages that name rivals but not the brand into a pitch list, and the brand's own cited pages into fixes. Use when the user asks whether ChatGPT, Gemini, Claude, Perplexity, Copilot or Google AI Overviews mention or recommend their brand, wants a GEO, AEO or LLM visibility check or AI share of voice, asks why AI recommends a competitor, or asks how to show up in AI answers. One run is a snapshot, not a trend.

1 other project uses a version of this.

Adopted by

Most recently taken up by

  • Other versions1 project uses a version of this.

    Word for word · 0

    None found.

    Edited · 0

    None found.

    Wrote their own version · 1

    1. peakanswer/peak-answer-skills★ 0same name

    How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.

  • In this repository5 other files it gives its agents.

The file

We split this skill into the rules it gives an agent

6 rules. About 0 are common, 4 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈4

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read references/ground-rules.md first.

    line 14
  2. Very specific:

    Write seo/prompts.md (template in the seo-geo-playbook skill, or make the table yourself): 20–25 questions a buyer would ask an AI before …

    line 18
  3. Distinctive:

    Use the AI models you can reach, with web search on: links in an answer come only from live retrieval (ANTHROPIC-WEBSEARCH).

    line 27
  4. Distinctive:

    Neither: ask the owner to paste answers from the apps, or use a tracker.

    line 31
  5. Distinctive:

    Ask each prompt at least 3 times per AI model, each run its own fresh session, never three turns of one chat: an earlier answer shapes the …

    line 40
  6. Distinctive:

    No page of the brand's own is cited, and the branded prompts ("what is {brand}") don't find its domain either.

    line 66

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AI Visibility Check

Finds out how AI answers treat a brand for its buyers' questions: named, recommended, cited, and who is named instead. You can't improve what you haven't measured, and you can't measure it by asking one question once.

Read references/ground-rules.md first. The keys in brackets are in references/sources.md.

1. The prompt set (once, then fixed)

Write seo/prompts.md (template in the seo-geo-playbook skill, or make the table yourself): 20–25 questions a buyer would ask an AI before buying, in their words, not as keywords.

  • Where they come from: sales and support questions, Search Console's longest queries, People Also Ask, Reddit threads, rivals' comparison pages.
  • The mix: problems ("how do I…"), the category ("best X for Y"), comparisons ("X vs Y", "alternatives to X"), and a few branded ones ("what is {brand}", "{brand} pricing", "{brand} vs {rival}", "{brand} reviews"), about one in seven, kept apart in every count. Branded prompts only reach people who already know the brand.
  • More prompts beat more runs of one: going from 10 to 100 prompts cut the spread of a brand's measured coverage by about 73% (OTTERLY-STABILITY).
  • Keep the set fixed so the weeks compare. Retire a prompt only with a reason in the ledger.

2. Ask

Use the AI models you can reach, with web search on: links in an answer come only from live retrieval (ANTHROPIC-WEBSEARCH).

  • API keys in the environment: OpenAI with web search, Gemini with Google Search grounding, Perplexity, Anthropic with the web search tool.
  • Only your own model with web search: that's one AI model. Label it as such.
  • Neither: ask the owner to paste answers from the apps, or use a tracker.

Stamp every run with the AI model, API or app, the date and whether it searched. API answers aren't app answers: Google plus OpenAI's API recover only about a quarter of the pages ChatGPT's app cites (JIANG), and forcing search on prompts that don't search on their own moved measured visibility by about 20 points (GRAPHITE-SEARCH).

Ask blind. Your own session knows the brand from this conversation, so its answers lean toward it. Run each prompt in a fresh session or a subagent that gets the prompt and nothing else: not the brand, not the site, not why it's asked. In an app, use a new chat with memory off. Never ask in the chat where the brand was discussed.

  • In Claude Code, a subagent isn't blind. It sees the project's skills, and a skill's description can name the brand. Run each prompt as its own headless session from an empty folder outside the project (claude -p). Claude Code also tells every session the signed-in account's email: when its domain points to the brand, count those runs apart. Both turned up in the author's run on sightkick.so on 2026-11-11.
  • Spread the runs over sessions. In that run, one Claude Code session stopped at 200 web searches, after about 33 runs.

Ask each prompt at least 3 times per AI model, each run its own fresh session, never three turns of one chat: an earlier answer shapes the next. Even 7 runs a day leave about ±16 points of error (SCHULTE). Say how many you ran.

3. Read each answer: four signals, apart

  • Named: the brand appears by name.
  • Recommended: the brand is suggested as a pick for the person asking.
  • Cited: the answer links to one of the brand's pages.
  • Sentiment: how it's described.

They disagree: in Semrush's data, 61.7% of brand appearances were a link with no name, ChatGPT cited without naming and Gemini named without linking (SEMRUSH-GHOST). Never merge them into one score. There is no standard share-of-voice formula, so never compare one tool's score with another's (DIGIDAY-TOOLS). Write down the rivals named and every cited URL, run by run, in seo/prompts.md.

4. Read the sources

Sort every cited URL:

  • The brand's own page: what does a rival page cited for the same prompt say that this one doesn't? That's a fix for seo-page-refresh. In matched groups, brands whose own pages were cited were named 40–49 points more often (TANNENBAUM).
  • A third-party list, review or comparison that names rivals but not the brand: the pitch list (step 5). Being on cited lists was the strongest of the factors one small vendor study tested for ChatGPT recommending a B2B software brand (r 0.64, 278 prompts, POSITION-LISTS, August 2026, before GPT-6).
  • A rival's own page: what it covers that the brand's site doesn't is a topic for seo-page-writer. ChatGPT's own searches lean on brand sites, with "official" and "price" among its rising search words (AHREFS-FANOUT).
  • YouTube, Reddit, LinkedIn, review sites: off-site options. Google's AI answers cite YouTube and Reddit heavily (AHREFS-AIO50). ChatGPT barely does: YouTube in about 2% of software answers (FOUNDATION-YT), Reddit in 1 answer in 43 by late September, before GPT-6 (AHREFS-FANOUT). A YouTube comparison that names the rivals and a founder's LinkedIn article are placements the brand can make itself (FOUNDATION-YT, SEMRUSH-LINKEDIN).

AI models barely share sources: in one October 2026 study, ChatGPT shared about 5% of its cited pages with Google's AI (TJR). Read each AI model on its own.

ChatGPT changed model on 2026-10-07. GPT-6 rolled out from that day, and linked pages per answer rose from 3.3 to 4.1 (SISTRIX-GPT6). It's a break in every ChatGPT trend: most ChatGPT figures in this set were measured before it, so read them with their dates. Mark the date on your own trend lines too, and compare ChatGPT weeks on the same side of it.

When AI can't find the brand at all

No page of the brand's own is cited, and the branded prompts ("what is {brand}") don't find its domain either. Then the answers can't be fixed page by page yet. Check in this order:

  1. Is the site in the index the AI model searches? Bing's index grounds Copilot: check Bing Webmaster Tools, or search Bing for the domain (BING-AIPERF). For Google's AI Overviews and AI Mode, check Search Console's page indexing report, or search Google for the domain. ChatGPT's search shows only sites that let OAI-SearchBot in (OPENAI-BOTS). A site missing from the index goes to seo-geo-audit first: nothing else helps until it's in.
  2. Is another brand answering to the name? Read what the answers say "{brand}" is. If they describe another product with the same name, every prompt reads as that one until the site and its profiles say plainly what the brand is: the same name with its category ("{brand}, the {category} for {buyer}") on the home page, the About page and every profile (seo-geo-audit, Gate 5). On the author's own SEO tool, in a first blind check with one AI model and web search (9 runs, 2026-11-09), no answer found its domain, and 3 of 9 described a drum app with the same name.
  3. Then the usual work: the pitch list (step 5) for third-party pages already cited for the buyer's questions, and pages of its own for the prompts where rivals are named (seo-page-writer).

5. The pitch list

Pages the AI answers already cite for the buyer's questions that name two or three rivals but not the brand.

  • Pick by fit, never by domain rating. Google doesn't use third-party scores like DR (MUELLER-DA). Never pitch a rival's own pages.
  • On a list, ask for a fair place, not just a mention. The top spot in a cited list went with 16.5 more visibility points (PEEC-LISTRANK).
  • One page per pitch: why this page, what the brand adds, and one fact the page is missing or gets wrong. 61–80 words, from an inbox on the brand's own domain. Three emails beat one (HUNTER-OUTREACH), and three is the most we send. Write it from the page's content, not from flattery (BUZZSTREAM-12K). One outreach lead says 8 of his last 14 replies agreed to a pitch that opened with a correction, with no baseline (SEMRUSH-OUTREACH).
  • Expect little: link builders get about 3.6% replies per email (BUZZSTREAM-REPLIES). The author's own outreach for an SEO tool in 2026: 81 pitches, 15 replies, 12 of them asking to be paid, no free links.
  • Never pay for a placement or swap links. Google requires a paid link to carry rel="sponsored" or nofollow and prefers sponsored, so use sponsored. A paid placement is the owner's call (GSP, G-QUALIFY).
  • Draft the pitches. The owner sends them, unless they've said you may.

6. Check again

After a change or a placement, ask the same prompts again, over weeks. Count weeks cited, not single citations. Citations blink: in Ahrefs' 20-day study, 50–59% of day-one citations on AI Overviews, AI Mode and ChatGPT were gone the next day, and about 75% of those that disappeared came back within the 20 days (AHREFS-BLINK, 2026-10-29). In Profound's sample, the median cited page lost half its citation share within 11 days of its peak (PROFOUND-HALFLIFE). Call a change only when it's bigger than the run-to-run spread (SIELINSKI).

What moves AI answers, and what doesn't

Doesn't: llms.txt, schema as a lever, chunking pages, "noai" tags (references/ground-rules.md, Not levers). Does: bots that can get in (seo-geo-audit), pages that name the entity, carry a date and say something new (seo-page-writer), and other sites that name the brand. In one 46-prompt ChatGPT study, brands on six or more pages cited for a question were about six times likelier to be recommended than brands on one (VISIBLE-PLACEMENTS).

Output

Open with the AI models that ran: each one's name, how it was asked (API, app, or your own model with web search), the date and the runs per prompt. With only your own model, say so in the first line ("1 AI model: Claude with web search, asked by this agent"), so nobody reads it as ChatGPT's answer.

| # | Prompt | AI model | Runs | Named | Recommended | Cited | Rivals named |
| - | - | - | - | - | - | - | - |

Then: if AI can't find the brand, that first, with what the index and name checks found; the weakest AI model, the biggest gap (a prompt where rivals are named and the brand isn't), the top 5 pitch targets, the top 3 own pages to improve, and the runs and dates. Say plainly that one sitting is a snapshot.

With Sightkick

One sitting is a snapshot. Sightkick asks the prompt set every day across AI models and keeps the history, and its free AI visibility report shows today's picture with no card. If it's connected, the sightkick skill maps each step to its call.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.