Key takeaways
- A handful of open-source repos on GitHub now track brand visibility in ChatGPT, Perplexity, and AI Overviews. GitHub's ai-visibility-tracking topic lists 26 public repos, so the field is wide but young.
- The biggest technical question is whether a tracker calls the ChatGPT API or scrapes the real ChatGPT interface. The API often lacks the web-search citations and ads users see.
- Single runs tell you little. ChatGPT answers vary, and Promptwatch Data shows 47.6% of cited responses name 10 or more unique domains, so you need many prompts and repeated runs.
- Self-hosting is cheap on licence cost but costs time: API keys, scheduling, storage, and maintenance are yours.
- Verify each repo yourself (licence, last commit, install method) before committing. I have not run these, and the details below come from repo descriptions.
What "open-source ChatGPT rank tracker" actually means
Traditional rank trackers record where a URL sits for a keyword. ChatGPT has no fixed page one. It writes an answer, names some brands, and cites a set of sources that changes between runs.
So a ChatGPT rank tracker does something different. It sends a portfolio of prompts to one or more AI engines on a schedule, stores the answers, and extracts three things: was my brand mentioned, where in the answer, and which URLs were cited.
The open-source version of that is a small app you run yourself. Typically you supply API keys, point it at a database, define your prompts and competitors, and let a cron job do the rest. If you want the broader picture of how AI answers get built before picking a tool, start with Promptwatch Data, which is first-party research on AI citations.
A quick clarification, since the search results for this topic are noisy: self-hosted LLM leaderboards and "self-hosted ChatGPT alternatives" like local chat UIs are a different category. Those help you run a model. They do not tell you how a model talks about your brand.
The open-source trackers worth looking at
The list below comes from GitHub repo descriptions. Treat features as claims until you read the README and try the install.
| Project | Stack / notes | Engines mentioned | Notable angle |
|---|---|---|---|
| Open-Source-SEO-Tools/ai-visibility-tracker | Self-hosted GEO visibility tracker | ChatGPT, Perplexity, AI Overviews | Brand visibility and citations |
| CiteLadder (abhij1306) | Open-source, self-hostable, DataForSEO and MCP topics | Multiple engines and markets | Share of voice, owned citation rate, fan-outs, AI ads, CSV export |
| GEO/AEO Tracker (danishashko) | MIT licence, TypeScript, Next.js, web scraping | ChatGPT, Gemini, Perplexity, Copilot | Positioned as an alternative to Semrush AI Visibility and AthenaHQ |
| OneGlanse | Open-source AEO/GEO tracker | Several AI engines | The model you connect scores each answer; dashboard shows rank, sentiment, sources |
| rastmob/ai-visibility-tracker | TypeScript, Next.js | Gemini, Claude, Perplexity, ChatGPT | Lightweight multi-engine tracker |
| logwick (stani-chirk) | Self-hosted server-log analytics | AI bots | Complement, not a rank tracker |
Open-Source-SEO-Tools ai-visibility-tracker
This is the most direct match for the keyword. It describes itself as a self-hosted tracker for brand visibility and citations across ChatGPT, Perplexity, and AI Overviews. Good starting point if you want the plainest possible scope.
CiteLadder
CiteLadder has the longest feature list of the group: scheduled audits of a prompt portfolio across engines and markets, visibility, share of voice, owned citation rate, average position, competitor ranking, cited URLs, accuracy fact checks, AI ads, and query fan-outs. The DataForSEO and MCP topics suggest it can pull from external data providers and plug into agent workflows. That is ambitious for a self-hosted project, so check how much is built versus planned.
GEO/AEO Tracker
MIT licensed, built on TypeScript and Next.js, and aimed squarely at teams who looked at paid tools and wanted a free one. The topics list web scraping, which hints at it reading real interfaces rather than only calling APIs. Confirm that in the code, because scraping logged-in chat UIs can break often and may conflict with a provider's terms.
OneGlanse and rastmob
OneGlanse uses the model you connect to score each answer, then shows visibility, rank, sentiment, recommendations, sources, and change over time. That is a sensible design, though it means your scoring quality depends on whichever model you wire in. rastmob's tracker is a leaner TypeScript and Next.js option covering four engines.
logwick, a different job
logwick does not measure rank. It analyses server logs to spot AI bots and agents hitting your site. That is useful next to a tracker: the tracker tells you what AI says, the logs tell you whether AI crawlers can reach your pages at all. Promptwatch Data's AI crawler traffic report shows how much verified crawler volume is out there to monitor.
API vs real ChatGPT: the decision that matters most
Most self-hosted trackers are cheap to build because they call model APIs. The catch is that an API response is not necessarily what a person sees in ChatGPT. The research notes I worked from flag that the API usually lacks the web-search citations and ads shown in the interface. I could not verify how large the gap is for every tool, so test it yourself: run ten prompts through the tracker, then ask ChatGPT the same ten by hand and compare cited sources.
This matters because the interface is changing. Promptwatch Data recorded ChatGPT starting to use the site: operator at scale on August 8, 2026, going from about 0.4% to about 17% of fan-out queries overnight (report). It also logged average citations per response dropping around the GPT-5.3 rollout on March 4, 2026 (report). A tracker pinned to an API model can miss that sort of shift entirely.
Design rules for any tracker you run
These come from how ChatGPT answers actually behave, and they apply whether you install a repo or build your own.
Track many prompts, not a few
Promptwatch Data found that for ChatGPT responses with citations, between Sep 28 and Oct 4, 2026, 47.6% cited 10 or more unique domains and 66.1% cited at least 8. Only 1% cited a single domain (report). With that many sources in play, one prompt is a lottery ticket. Build a portfolio of dozens to hundreds of prompts, grouped by topic.
Repeat runs
Answers vary between identical runs. Run each prompt several times and report rates (mentioned in 6 of 10 runs) instead of yes or no.
Log fan-outs if the tool supports it
ChatGPT turns a prompt into several web searches. Promptwatch's query fan-out data shows average fan-out length dropping from about 117 characters in December to roughly 53 characters from April 2026. Tools that store the actual sub-queries, as CiteLadder claims to, help you write headings that match how the model searches.
Track position and wording, not just presence
Brand-specific prompts are different from open ones. In the same report, 21.3% of brand-specific responses cited 10 or more domains, with five domains the most common count. Segment by prompt type so branded and unbranded results do not blur together.
Keep a change log
Write down model updates and site changes next to your charts. Without annotations, a visibility drop after a model rollout looks like a content problem.
Setting up a self-hosted tracker: a realistic checklist
- Read the README for licence, install method (Docker or manual), and required keys.
- Check the last commit date and open issues. A repo untouched for months is a risk when AI interfaces change monthly.
- Estimate cost: prompts x engines x runs x days. Multiply by your API price, then add scoring calls if the tool uses an LLM to grade answers.
- Start with 30 to 50 prompts and one engine. Expand once the numbers look stable.
- Export to CSV or a database you control. Pipe it into a BI tool such as Metabase or Grafana if you want dashboards.
- Schedule runs and alert on failures. A silent cron failure produces gaps nobody notices for weeks.
A quick example of a minimal Docker-based setup pattern many of these repos follow:
git clone https://github.com/<owner>/<repo>.git
cd <repo>
cp .env.example .env # add your API keys and database URL
docker compose up -d
Check each project's own instructions, since commands differ.
For workflow glue, n8n is itself open-source and self-hostable, and works well for scheduling prompt runs and sending Slack alerts around a simple tracker.
Where self-hosting wins, and where it does not
Self-hosting makes sense when you want full data ownership, have a developer who enjoys the project, and can tolerate rough edges. It also suits a small prompt set where API costs stay low.
It gets painful when you need multiple countries and languages, real-interface data, crawler logs, traffic attribution, or reports clients can read. Those are the parts open-source repos mostly skip, because they are tedious to build and maintain.
Here is how the options compare at a glance:
| Approach | Licence cost | Setup effort | Real UI data | Crawler logs / traffic | Content workflow |
|---|---|---|---|---|---|
| Open-source repo | Free | High | Depends on repo | Rarely | None |
| Light hosted tracker | Low monthly | Low | Varies | Usually no | Usually no |
| Full AI visibility platform | Higher monthly | Low | Yes | Often yes | Often yes |
Hosted trackers worth comparing
If you would rather not maintain code, several hosted tools cover the same ground:
Otterly.AI and Peec AI are straightforward prompt trackers.
Otterly.AI
LLMrefs and Rankshift track visibility across several engines.
LLMrefs
xSeek adds GPTBot crawler monitoring to ChatGPT rank tracking.
For a wider sweep, the AI rank tracker directory at ai-rank-tools.com and the GEO software list at bestgeosoftware.com are good places to keep browsing.
When you need the full stack
If the goal is not just measuring but fixing, a platform like Promptwatch is the sort of thing to look at. It monitors the real interfaces of ChatGPT, Gemini, Perplexity, Claude, and AI Overviews rather than only APIs, adds IP-verified AI crawler logs and visitor analytics, and includes a Content Agent that writes and publishes pages against the gaps it finds. That covers the pieces an open-source tracker usually leaves to you.
I would not call it a replacement for tinkering. If you like owning the stack, a repo is a fine learning project. If you are reporting AI visibility to a boss or a client, the maintenance burden of a homemade tracker catches up fast.
Pitfalls to expect
- Model changes break comparisons. After the GPT-5.3 rollout, citations per response dropped across models, which would have looked like a ranking collapse in a naive dashboard.
- Scraping fragility. Trackers that drive a browser break when the interface changes, and may violate terms of service.
- Scoring bias. Tools that use an LLM to judge sentiment or accuracy inherit that model's quirks. Spot check a sample by hand.
- False precision. A visibility score of 63.4 means little when the underlying prompt set is 20 prompts run once.
- Source mix shifts. Promptwatch Data recorded reddit.com's share of ChatGPT citations falling from about 4% to 0.5% on August 14, 2026 (report). A tracker should let you see which domains are cited, not just whether you are.
A sensible path
Start with one of the repos above on a small prompt set, and compare its output against manual ChatGPT checks for a week. If the numbers match and the project is actively maintained, keep going. If the API-versus-interface gap is large, or you find yourself writing glue code for crawler data, traffic attribution, and reports, that is the signal to price a hosted option.
Whichever you choose, the habit matters more than the tool: a fixed prompt portfolio, repeated runs, annotated model changes, and a monthly review of which pages actually get cited.