Do AI assistants recommend you or your rival?

What it is and why it matters

When someone asks ChatGPT “what's the best app for…?”, the answer names a few products and often cites some sites as sources. If your name isn't there, the reader never sees you. And you don't know, because nobody tells you.

Limelit Open measures exactly that. You give it your site and up to five rivals, and it asks ChatGPT, Claude, Perplexity and Gemini a customer's questions, through their APIs and with web search on. Then it counts how often you come up and how often your rivals do, in what position, and which sites get cited. All in a dashboard on your computer, which you can query from Claude Code.

Two things to know right away. It's a third-party open-source project, not mine, under the Apache-2.0 license. And it's still pre-release, meaning a trial phase: there's no final version. It's for anyone with an app, a SaaS or a website: indie hackers, solo founders, vibecoders.

1. What it measures, and what it doesn't

Each pass saves the full answers and looks for your name and your rivals' in them with a plain text search: no model decides what it saw, so every number can be checked by hand. The numbers:

• Visibility: the share of answers that name you.
• Share of voice: your mentions out of those of all the brands you track.
• Position: your average place when the answer is a list.
• Citation share: how many cited sources are your site. Every cited site is classified: yours, a rival's, social, informational, other.
• The searches the engine ran before answering, which often aren't the question you asked.

Every number says how many answers it rests on (n), and under 20 the dashboard flags it. Questions that name your brand are left out of the headline number: they would only measure the question.

The most important limit

The answers come from the models' APIs, with web search on, and they aren't identical to what a person sees in the ChatGPT app: the README says so plainly. To measure what people see you need scrapers, paid services that read the page the way a user does. The same goes for Google's AI Overview and AI Mode and for Bing Copilot, which have no API: either a scraper or Limelit's hosted version. Take the numbers as a direction.

2. Install it and run the first pass

It runs on Mac or Linux, it's a single file, and it creates its own SQLite database. Nothing to install first. Open the Terminal and paste:

mkdir -p ~/limelit && cd ~/limelit
curl -fsSL https://github.com/limelit-co/open/releases/latest/download/limelit_$(uname -s | tr A-Z a-z)_$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/').tar.gz | tar xz limelit
./limelit serve

1. Open http://localhost:1515 in your browser. The Terminal also prints ready-made lines to connect Claude Code, with your paths.
2. The setup wizard asks for your domain, then the category, meaning the phrase a customer would use (for example “app to learn languages”), and up to five rivals.
3. It suggests starter questions, in English. Untick the ones your customer would never ask and add your own below, one per line, in your language too.
4. Last step: a key to reach the engines. There are two ways, covered in section 3.
5. Press Run: nothing is spent before that button. The answers usually arrive within a minute.

My advice: the questions matter more than anything else. Write them the way a customer who doesn't know you would: what they want to do, for whom, with what constraint. A starting point to adapt:

What's the best app for [what your app does]?
Which app do you recommend for [problem] if [constraint, for example: I don't have much time]?
Is there a free app for [what your app does]?
[Rival] alternatives?
How do I choose an app for [what your app does]?

3. The key: the free one or your own

The wizard asks you for a key, the code that lets you use the engines' APIs. Two ways:

• The free key. You sign in with Google to Limelit Cloud, create the key and paste it into the wizard (or run limelit login). It reaches ChatGPT, Gemini, Perplexity and Google's AI Overview and AI Mode, within a free monthly allowance. But your questions pass through Limelit Cloud's servers to reach the engines; the answers stay with you.
• Your own keys. The quickest is OpenAI's, which opens ChatGPT; for Claude you need Anthropic's. In Settings you'll find every provider with a link to the page where the key is created. You pay the providers directly, at their rates: the project adds no costs of its own.

With your own keys there's a ceiling, runs_per_day, which you change in Settings: a pass that would exceed it is refused before anything is spent. Limelit also counts calls and tokens per engine per day.

If you leave limelit serve running, it repeats the pass on its own every day or every hour: that's how you see who's winning day by day.

4. Connect it to Claude Code

Limelit Open has an MCP server, a doorway that lets Claude read your data and answer you in plain words, with the same numbers as the dashboard. In the Terminal:

claude mcp add limelit -s user -- sh -c 'cd ~/limelit && exec ./limelit mcp'

-s user makes it available in all your projects. It has to start from ~/limelit to read the same database as the dashboard: started anywhere else, it opens an empty one. Then open a new Claude Code session and type /mcp: limelit should show as connected.

You don't need to know the tool names. Start here:

Let's get started with Limelit: where does my install stand and what should I do next?

Claude tells you where you stand and what to do next. It can also start a pass: first it shows you the plan, and it only starts when you say yes.

Or with the Limelit plugin

The repository is also a Claude plugin marketplace. In Claude Code, type:

/plugin marketplace add limelit-co/open
/plugin install limelit-open@limelit

limelit-open connects your install with the skill limelit-open-visibility-report. There's also limelit-audit, with no install and no account: ai-crawler-check tells you which AI crawlers your robots.txt lets in, llms-txt-check audits your llms.txt. They may send the domain to Limelit's free checker.

5. The questions to ask Claude Code about your rivals

Once it's connected, these are the questions I recommend. Replace the parts in [ ] with your own.

Which questions am I losing? For each one, tell me which rival gets named instead of me, on which engine, and how many answers the number rests on.

In the answers where I don't appear, which sites get cited? Give me the exact page addresses, ordered by how often they're cited.

For the question “[question]”, what searches did the engine run before answering? Which words does it use that I don't use on my site?

Compare me with [rival]: which questions they win, which ones I win, and which sites the answers they win cite. If the sample is small, tell me before drawing conclusions.

6. What to do with the results

These are my own tips, not Limelit features: it measures, you do the work.

• Get onto the cited pages. If the answers keep citing the same “best apps for…” ranking, that page counts more than your site: write to whoever runs it and present your app.
• Write the page that answers the question. If you're losing on “is there a free app for…?”, make a page that answers exactly that, with precise facts.
• Make an honest comparison page. For “[rival] alternatives”, explain when it's worth choosing you and when it isn't.
• Use the engine's words. If its searches use different words from yours, put them in your page titles.
• Measure again after a few weeks, not after a day: answers change on their own too.

Common mistakes

• Filling the list with questions that name your brand. You need questions from a customer who doesn't know you.
• Drawing conclusions from a few passes. Look at n: under 20 answers it's a small sample, not a trend.
• Forgetting that with the free key your questions go through Limelit Cloud. If that doesn't suit you, use your own keys.

Checklist

• limelit serve is running and the dashboard opens at localhost:1515
• The questions sound like a customer, and I've added my own in my language
• I've chosen the key: the free one (it goes through Limelit Cloud) or my own
• I've checked runs_per_day before turning on the automatic pass
• In Claude Code, /mcp shows limelit as connected

Frequently asked questions