Weekly visibility for indie software

Mentioned is not recommended.

AI answers can name your product and still send the buyer to someone else. Indie AEO Lite repeats a fixed set of buyer prompts and scores mention rate and recommendation rate separately, including the range across those repeats.

Open the dashboard Early access

Fictional Northwind Notes mock sample: mention stays 60.0% (12/20); recommend moves from 40.0% to 20.0%, shown as 20.0–40.0%. Pilot Pad is recommended more often on the later pass. Not a model log and not a ranking.

One sentence, two scores

Scored by the same rules as a run. This sentence is an illustration, not a model log.

Northwind Notes comes up in roundups, but I would not recommend it. Pilot Pad is the top pick.

Northwind Notes

Mention only

The word “recommend” is negated.

Pilot Pad

Recommended

Cue: top pick.

A rate is the share of prompts in a run, not a badge on one sentence. Mention rate is mentions divided by prompts. Recommendation rate is recommendations divided by prompts.

What a week looks like

01Save a brand, up to five competitors, and a fixed set of buyer prompts. Twenty starters are included. You can print that set for an audit.
02Run the same prompts at least twice, on the same provider, model, and temperature. Mock needs no key. OpenAI uses Chat Completions when you add one.
03Rules tag every answer: did it mention the brand, and did the same sentence recommend it?
04The digest shows mention and recommend, the min–max across repeats, who was recommended more than you, and one next step. With no email key, it is a markdown file in exports/.

How the rates are counted

The same formulas the probe writes onto a completed run.

mention_rate = answers that mention you / prompts
recommendation_rate = answers that recommend you / prompts
gap = mention_rate - recommendation_rate
range = min and max of those rates across repeats

A share sort (recommendation count, then mention
count, then name) can sit in a table. It is not
an AI search position.

What this does not know

There is no access to official ChatGPT, Perplexity, or Google logs. A run stores the text the provider just returned.
Scores are a sample plus a rule scorer. A thin sample is not an SEO ranking and not a position in AI search.
The scorer is keywords and substrings. “Acme is popular. I recommend it.” counts as a mention only, because the cue is in the next sentence.
Negation is shallow and English-only. A short name can match inside another word.
The mock provider ignores the prompt and cycles five scripted answers. A later repeat swaps brand recommendations for a competitor squeeze so you can see a range without an API key. That swap is the script, not model noise.
Google AI Overview is not included. This app does not scrape Google.

Run it locally

In development the dashboard opens as dev@localhost. Set DATABASE_URL first. npm run probe uses that database, creates the fictional Northwind Notes brand when it is empty, then writes metrics.

npm install
cp .env.example .env.local
npm run dev
npm test
npm run probe