People ask ChatGPT for an app, and they search Google and land on its website. The spec records both next to the listing. It runs on your own model keys and your own Search Console, and nothing needs an Appeeky account.
Two more ways to be found
A store search is one way in. An assistant answering “what app helps me sleep” is another, and it names three or four apps, not a page of results. The website is the third: it ranks in Google, and it is one of the pages the assistants read and cite.
The spec keeps all three as files. That way your agent reads them before it proposes a change, and verify measures them after each release.
Your keys, your bill
- ChatGPT
OPENAI_API_KEYgpt-4.1 with web search - Claude
ANTHROPIC_API_KEYClaude Haiku with web search - Gemini
GEMINI_API_KEYGemini 2.5 Flash with Google Search - Perplexity
PERPLEXITY_API_KEYSonar, which always searches - Search Console
GSC_CREDENTIALSa service account key, or your gcloud login - Google ranks
DATAFORSEO_LOGINwith DATAFORSEO_PASSWORD, your DataForSEO account
Any of them. Each model key adds that assistant to the scan, and that provider bills you directly. With an Appeeky plan, Appeeky AI Visibility can run the scans for you instead and write the same file. Pick one per repo.
What people ask
An intent is what someone wants. Its prompts are how they would ask an assistant. Both live in the spec. To start, have your model write them from the listing:
$ npx @appeeky/spec visibility init --local✓ US: wrote 6 intents to aiVisibility.intents in appeeky.spec.yamlgpt-4.1-mini on your key, about $0.004. Edit them, then: appeeky-spec visibility scan --local --dry-run
# appeeky.spec.yamlweb:url: https://sleepsounds.app/aiVisibility:countries: [us]models: [chatgpt, claude, gemini, perplexity] # optional: the ones with a keyaliases: [Sleep Sounds] # other names the assistants useintents:us:- label: Fall asleep to rain soundsprompts:- what app plays rain sounds to help me fall asleep- best free rain sounds app for iphone
Edit them like any other part of the spec. A prompt sounds like a person typing. It never contains the app's own name, and it uses the storefront's language.
Ask the assistants
$ npx @appeeky/spec visibility scan --local --dry-runUS 6 intents · 24 prompts × chatgpt, claude, gemini, perplexity · about $2.45About $2.45 on your own keys in all. Nothing was asked.
Without --dry-run, it asks after you confirm. Each prompt goes to each assistant with web search on, the way the apps answer people. A second, cheap call then lists the apps each answer names, in order. A provider that fails is reported, and the rest of the scan goes on.
Read the answers
# appeeky/visibility/us.yamlsource: localsummary: { winning: 2, weak: 1, missing: 3, noData: 0 }intents:- label: Fall asleep to rain soundsstatus: missingyou: 0byModel: { chatgpt: 0, claude: 0, gemini: 0, perplexity: 0 }namedInstead: [Calm (75%), BetterSleep (50%), Rain Rain (25%)]sources:- { domain: reddit.com, count: 6, url: https://www.reddit.com/r/sleep/… }- { domain: sleepsounds.app, count: 2, yours: true, url: https://sleepsounds.app/rain }
missing means no assistant names the app for that intent. namedInstead is who they name. sources lists the pages they read, and yours: true marks your own site. Assistants repeat what the web says. If they cite review sites and Reddit threads that name your competitors, those pages and your own website are what to work on, along with the listing's first lines.
The site, as a crawler sees it
$ npx @appeeky/spec seo checkSearch✓ Title: “Sleep Sounds: rain and white noise” (34 characters)! No canonical linkThe app! No Smart App BannerAI assistants✗ robots.txt blocks GPTBot, ClaudeBot! No llms.txt! 86 words readable without JavaScript12 of 18 fine.
It needs no key. It reads the page, robots.txt, the sitemap and llms.txt, and every finding comes with its fix. The AI part matters most: most AI crawlers do not run JavaScript. A blocked crawler, or a page that only renders in the browser, leaves the assistants with nothing to cite.
$ npx @appeeky/spec seo llms --out public/llms.txt✓ public/llms.txt
llms.txt is written from the listing: what the app is, where to get it, the searches it is for, and its description. Run it in your website's repo, or have your agent propose it there.
Google Search Console in the repo
- In Google Cloud, create a service account and a JSON key, and turn on the Search Console API.
- In Search Console, add the service account's email as a user of your property.
- Set
GSC_CREDENTIALSto the JSON, or to its path.
$ npx @appeeky/spec seo pull✓ appeeky/seo/ sc-domain:sleepsounds.applast 28 days: 1,240 clicks (+18%) · 41,900 impressions250 queries, 64 pages. Read appeeky/seo/summary.md.
daily.csv keeps clicks and impressions per day across pulls. queries.csv and pages.csv cover the last 28 days. summary.md picks out the queries close to the top (position 4 to 15), where a better title or section can move them onto the first screen, and the queries that use your store targets. A phrase people type into Google is often what they ask an assistant too, and it can make a good target.
Google ranks on DataForSEO
Search Console shows the searches you already appear for. To see the searches you want to win, use your own DataForSEO account. It checks Google for each keyword in web.keywords (or your store targets), the way a phone sees the results.
$ npx @appeeky/spec seo ranks6 keywords in US · about $0.12 on DataForSEO✓ appeeky/seo/ranks.md 3 of 6 in the top 10 · AI Overview cites you 2×US rain sounds site 6 store page 2 AI cites youUS white noise app site 14 store page — AI cites calm.com, reddit.com
Each keyword gets three answers. Where your site ranks. Where your App Store or Google Play page ranks on Google, which is often higher than the site. Whether Google's AI Overview shows, and if it cites you. A keyword that enters or leaves the top 10 goes to changes.md.
seo keywords adds search volume, difficulty and related ideas. seo domain shows everything the site ranks for and which sites compete for the same searches. Every command quotes first with --dry-run. The daily routine checks ranks once a week when the account is set.
Measured with every release
$ npx @appeeky/spec verify2026-11-03 day 14: won 3 up · 0 downbetter AI answers that name the app 0.18 → 0.31better Google search clicks/day 38 → 46better average Google position of the site 18.4 → 11.2
verify reads ai_visibility from history.csv, and search_clicks and search_impressions from daily.csv, and web_rank and ai_overview_citations from ranks.csv. Assistants change their answers slowly and Search Console lags two days, so read both on day 14. The verdict goes into the playbook with the ranks, and the next proposal starts from it.