AI Search Visibility & GEO Audit + Coding Agent Fix Prompt
Pricing
$9,000.00 / 1,000 snoffle reports
AI Search Visibility & GEO Audit + Coding Agent Fix Prompt
AI search visibility and GEO audit for one product page: can AI read it, do buyers asking Google and AI find it, and who they find instead. Returns the score and a handoff prompt your coding agent runs to apply every fix. $9 per complete report, free quick check, no report no charge.
Pricing
$9,000.00 / 1,000 snoffle reports
Rating
0.0
(0)
Developer
Broke to Built
Maintained by CommunityActor stats
0
Bookmarked
2
Total users
1
Monthly active users
2 days ago
Last modified
Features
Share
Snoffle: AI search visibility and GEO audit that hands your coding agent the fixes
Snoffle checks one product page and returns a work order your coding agent can run. It reads the
page the way an AI assistant does, asks Google and an AI assistant the questions a buyer would ask,
checks whether other sites mention the product, then asks an AI buyer's agent whether it would buy.
The record it returns ends in handoff: a prompt you pass straight to Claude Code, Cursor,
Lovable, Bolt or Replit, with every fix in order and a check for each one.
Built for agents and for the solo makers who run them: an agent can call it through the Apify API or Apify's MCP server, read two fields, and hand the rest to a coding agent without a human in the loop.
The agent loop
- Your agent runs Snoffle on a URL (one call, about 90 seconds).
- It reads
scores.overallandverdict.decisionto decide whether the page needs work. - It passes
handoffto a coding agent, which applies the fixes and verifies each one. - After the deploy, it runs Snoffle again with
"fresh": trueto measure the change.
Two different problems, told apart
A crawler check answers "can AI read the page". Snoffle reports that, and the harder problem behind it, as separate scores:
- AI can't read you (
scores.readinessis low): robots.txt blocks AI crawlers, the offer only appears after scripts run, the price or the buy path is missing, no llms.txt or JSON-LD. These are fixes on your own site, andhandoffcovers them. - AI reads you but doesn't pick you (
readinessis high,foundis low): the page is fine, but buyers asking Google and an AI assistant are shown other products. The record names those products (foundInstead) and, where Snoffle could judge them, what they have that you don't (strengths).
Snoffle's own Store listing was the second kind on 2026-11-06: readiness 94, found 18 (example below).
The Snoffle score
scores.overall is the Snoffle score, 0-100: how likely buyers and AI assistants are to find the
product. It is a measure of visibility, not a forecast of revenue. It has three parts:
- AI can read it (
scores.readiness): can an AI tool read the page, the offer, the price and the files without running scripts. - Buyers find it (
scores.found): does the product come up when buyers ask Google and an AI assistant their questions. A search for the product's own name is reported on its own line and does not count here. - Others mention it (
scores.traction): do other sites mention it.
While buyers don't find it and nobody mentions it, the Snoffle score is capped at 40: a site can be perfectly readable and still sell nothing if nobody can find it.
On a phone (scores.visitors) is reported separately: what a person on a phone gets, from
Google PageSpeed Insights (mobile). It is never part of the Snoffle score.
What one report contains
| Field | What it is |
|---|---|
url | The page that was checked |
product | The product name Snoffle read from the page |
mode | How the report was made: full or judged (a charged report is never rules) |
scores.overall | The Snoffle score, 0-100 (above) |
scores.readiness | AI can read it, 0-100 |
scores.found | Buyers find it, 0-100 (null if it could not be measured) |
scores.traction | Others mention it, 0-100 (null if it could not be measured) |
scores.visitors | On a phone, 0-100 (null if it could not be measured) |
scores.projectedOverall | The Snoffle score after every listed fix, capped at 40 while buyers don't find it and nobody mentions it |
scores.pillars | Internal detail behind readiness; read the three parts above instead |
verdict.decision | What the buyer's agent decided: buy, recommend_to_human or skip |
verdict.headline | One sentence on why |
verdict.blockers, verdict.scamFlags | What stopped it, and anything that looked unsafe |
topFixes | The fixes worth the most points: title, gain (points), evidence (what Snoffle saw) |
fixesTotal | How many fixes the full report lists |
allFixes | Every fix, ranked: title, impact, effort (minutes, hours or days), gain, evidence |
foundInstead | The products buyers are shown instead of you: name, url, timesShown, strengths |
buyerQuestions | The searches Snoffle asked as a buyer |
honestCall | Whether it is worth pushing (worth), why, the angle, and the firstPage to write |
nameCheck | Products whose name clashes with yours, and suggestions for a clearer name |
money | What the gap costs you, in plain dollars |
understood | What an AI understood: what it is, who it is for, the price and how to buy, each with clear |
contradictions, phoneProblems, nextStep | Where your pages disagree, what breaks on a phone, and the one thing to do first |
verdict.reasons, verdict.wouldChangeMind | Why the buyer's agent decided, and what would change its mind |
handoff | The paste-ready prompt for your coding agent, every fix in order, each with a check |
reportUrl | The full report as JSON: every check, its evidence, the search results, rivals |
checkedAt | When the check ran |
fromCache | true when this is your own saved report from the last 12 hours (not charged); see Repeat runs |
charged | true on a paid report, false on a run that was not charged |
A real example (this listing)
We ran Snoffle on this Actor's own Store listing, https://apify.com/eliai/snoffle-report, on
2026-11-06, before we applied its fixes:
scores.overall40:readiness94,found18,traction8verdict.decisionskip: "The listing reads well (94/100), but store search doesn't find it: found 18/100 for the searches buyers type."- Top fix: rank for "ai search visibility checker" in Apify Store search, +33 points, evidence "Apify Store search: not in the 85 results it returned."
- 7 fixes in total, and a
handoffprompt that starts:You are improving the Apify Store listing of eliai/snoffle-report ... so buyers and AI agents searching the Apify Store find it and choose it.
This README's first heading and the prefilled input are those fixes, applied. The prefilled input returns that record for free (see Input).
Why it is worth running
- It answers with evidence, not opinion. Every fix says what Snoffle saw on your page, for example "/robots.txt answers 200 with the homepage HTML, so crawlers get no rules."
- Your coding agent does the work. The
handoffprompt gives it each change in order, with the file to create and a command to verify it. - It checks the buying step too. It looks at whether an agent can see a price, a way to buy and a path that works without a login, and it reads your llms.txt, robots.txt, sitemap, JSON-LD and agent card.
Why $9, when a page check costs cents
A cents-per-page checker gives you a list of problems. Snoffle gives you the finished work order:
the searches buyers actually type and whether you came up, a buyer's-agent verdict on whether it
would buy, and a handoff prompt your coding agent runs to fix everything, each step with its own
check. You pay once, only when that full report is in hand.
Need a cheap check on many URLs?
Snoffle is one deep report per run. For screening, use one of these first and run Snoffle on the pages that matter:
- The free check on this Actor (
"quick": true, see Input): scores and top problems in a few seconds, $0, 3 a day per Apify account. - AI Readability Checker, our crawler-access audit: robots.txt rules per AI crawler, llms.txt and the page as each bot sees it, $0.05 per site. It answers "can AI read it" only, not "do buyers find it".
Input
{ "url": "https://your-product.com/" }
One URL per run. A bare domain (your-product.com) works. Local addresses, IP addresses and
non-web URLs are refused before anything runs. The prefilled URL is this Actor's own listing: it
returns the real report of it above as a free sample record (example: true) without running a
check, so a first click on Start shows real output. Any form of example.com, example.org or
example.net returns the same free sample. Neither is charged.
Optional: "fresh": true runs a new check even if you checked the same URL in the last 12 hours
(use it after you deploy fixes). A fresh check is a new report and is charged.
Optional: "quick": true runs the free check instead, the same free check as on Snoffle's
own site: the scores, the headline and the top problems in plain words, from Snoffle's plain
rules, in a few seconds. It is free ("free": true, "charged": false), 3 a day per Apify
account. It has no buyer's-agent model, no Google search, no fix evidence and no handoff
prompt, so the full report can score differently. Use it to screen URLs, then run the full report
on the ones that matter.
Use it from an AI agent
- Apify API, start the run and fetch the record (the reliable way):
POST https://api.apify.com/v2/acts/eliai~snoffle-report/runs?token=YOUR_TOKEN&timeout=360&waitForFinish=60with body{"url": "https://your-product.com/"}. PollGET https://api.apify.com/v2/actor-runs/RUN_ID?waitForFinish=60untilstatusisSUCCEEDED, then read the record fromGET https://api.apify.com/v2/datasets/DEFAULT_DATASET_ID/items. - Apify API, synchronous (one call, returns the record):
POST https://api.apify.com/v2/acts/eliai~snoffle-report/run-sync-get-dataset-items?token=YOUR_TOKEN&timeout=360. Apify's synchronous endpoints wait at most 300 seconds. A report usually takes about 90, but a slow one can run past that: the call then answers HTTP 408 while the run carries on, and is charged if it finishes. If that happens, fetch the run's dataset as above instead of starting a new run. - Apify MCP server: add
eliai/snoffle-reportathttps://mcp.apify.comand your agent can call it as a tool. - Read
scores.overallandverdict.decisionfirst, then handhandoffto the coding agent.
Pricing
Pay per event: $9 per report, the same price as one report on Snoffle's own site. You are
charged once, only after a complete report is in hand: the Snoffle score and all three scores it
is made of (readiness, found, traction), a verdict, the top fixes and handoff. If one of
those scores could not be measured, you get what was measured with "incomplete": true, and the
run is not charged. A repeat of the same URL by the same Apify user inside 12 hours is not charged.
Free check: $0 ("quick": true), 3 a day per Apify account. Never charged.
Not charged: a bad or missing URL, a page Snoffle cannot load, a timeout, any error on Snoffle's side, a report missing its top fixes, or a report made without Snoffle's buyer's-agent model (when that model is unavailable Snoffle can only run its plain rules, which is not the full report). Those runs end green with a record that says what went wrong and "This run was not charged." If the run's maximum cost is set below $9, or its timeout is too short for a full report, the Actor stops before checking anything.
Repeat runs: Snoffle keeps your report for a URL for 12 hours, for the same Apify user. If you
ask again for the same URL inside that window, you get the same report back with fromCache: true
(Snoffle's cached flag) and are not charged, so a retry after a client timeout costs nothing.
Another Apify user's run of the same URL never returns your report; they get their own. To have
your fixes counted, run with "fresh": true (charged), or run again once 12 hours have passed.
Speed and limits
- A report takes about 90 seconds (a full run of this Actor on 2026-10-31: 81 s) and can take up to about 150 s when Google PageSpeed is slow. Set the run timeout to 360 seconds or more; the Actor refuses to start a check that the timeout could cut off.
- One URL per run. Snoffle checks the public page as a visitor sees it: no login, no form submissions, no payments.
- Search results and AI answers change from run to run, so
foundandtractioncan move on an unchanged site. Each record lists the exact questions it asked (buyerQuestions) and who came up instead (foundInstead), so you can see what a score is based on. Compare two runs by those fields, not by a few points of score.
FAQ
What does "Buyers find it" measure? Snoffle writes the questions a buyer would ask about your kind of product, asks Google and an AI assistant, and records whether your site appears. 0 means it did not appear for any of them. Searching your product's name does not count: a buyer who already knows the name is not the one you need to reach.
Why is my projected score 40 or lower after every fix? The fixes on your site raise "AI can read it". "Buyers find it" and "Others mention it" are held at what was measured, and while both are low the Snoffle score is capped at 40. Getting found and mentioned is what lifts it past that.
Is this an SEO audit? Partly. It covers what search engines and AI assistants need to read the page (title, description, JSON-LD, robots.txt, sitemap, llms.txt), and adds what an AI buyer's agent needs to buy: a clear offer, a price, a path that works.
Who makes it? Broke to Built. Questions: open an issue on this Actor's Issues tab.
Changelog
- 0.7 (2026-11-16):
- README and listing rewritten for agents: the agent loop up front, "AI can't read you" told apart from "AI reads you but doesn't pick you", and a plain note that search and AI answers vary between runs. The Snoffle score is described as visibility, not a revenue forecast.
- New section pointing to cheaper screening: the free check, and our AI Readability Checker at $0.05 per site. Price, input and output are unchanged.
- 0.6 (2026-11-08):
- One report is now $9 (was $29), the same as one report on Snoffle's own site.
- The $0.49 quick check is gone:
"quick": trueis now the free check, the same as on Snoffle's site, free, 3 a day per Apify account. - A full report missing one of its scores (
readiness,foundortraction) is no longer charged. The record still shows what was measured, with"incomplete": trueand the reason.
- 0.5 (2026-11-07):
- 2026-11-08: a quick check missing one of its three scores is no longer charged. The record
still shows what was measured, with
"incomplete": trueand the reason. - A paid report now carries the whole report in the record: every fix, who buyers find instead, the buyer questions, the honest call, a name check, the money line, what an AI understood, and the one next step. Before, the record held only the top 3 fixes.
- New
quickinput: a $0.49 quick check (scores, verdict and fix titles from Snoffle's plain rules, no model, nohandoff). The full report is unchanged.
- 2026-11-08: a quick check missing one of its three scores is no longer charged. The record
still shows what was measured, with
- 0.3 (2026-11-06):
- Snoffle checked its own listing and we applied the fixes: the title, description and this README now use the words buyers search (AI search visibility checker, product page SEO audit, AI agent readiness score), and say who it is for and why it costs $29.
- The prefilled input is now this listing's URL, and its free sample is the real report Snoffle made of this listing on 2026-11-06, not a placeholder. example.com still returns it for free.
- 0.2 (2026-11-01):
- Each Apify user gets their own 12-hour saved copy: the Actor sends your Apify user id to
Snoffle, so another user's run of the same URL never returns your report, and yours is never
theirs. New optional input
freshruns a new check inside the 12 hours. scores.gradeis no longer read: Snoffle deprecated it. The record carries it only while Snoffle still sends it, and the status line and the dataset view no longer show it. The record addsscores.traction(others mention it) andscores.visitors(on a phone).scores.projectedOverallis now the Snoffle score after every listed fix, capped at 40 while buyers don't find it and nobody mentions it (it was a different, older score after the top 3 fixes only). The free sample record matched the example of that version.- Timeouts: the Actor waits up to 300 s for Snoffle, refuses to start (or to retry) with less than 330 s of run time left, and bounds every step after the answer, so a slow report is never cut off mid-run.
- A report Snoffle made without its buyer's-agent model (
mode: "rules"), or one missing its top fixes, is no longer charged. The record now carriesmode. After a timeout the Actor no longer retries (that would start a second full check). - After the charge, the Actor tells Snoffle which run was charged and at what price, so Snoffle's books follow what Apify charged. Your record and your charge are unaffected by it.
- A run that Apify restarts (a platform migration) after it already delivered and charged the report stops there instead of writing a second record.
- Every form of example.com (www., any path) returns the free sample. The run's output links
only the dataset; the full-report link is the record's
reportUrl, set only when a report was produced.
- Each Apify user gets their own 12-hour saved copy: the Actor sends your Apify user id to
Snoffle, so another user's run of the same URL never returns your report, and yours is never
theirs. New optional input
- 0.1 (2026-10-31): first version. One report per run, charged only on a complete report.