AI search testing infrastructure · open source
Unit tests for
AI search.
See who ChatGPT, Perplexity and Google AI Overviews cite instead of you, why, and fail the build when a change would cost you a citation. Free, offline, no API key needed.
Add it as a custom connector, sign in with one click, then ask “Audit my site for AI search”. See it in a chat
The six pillars in 88 seconds
Inside Claude, ChatGPT and your editor
Ask in plain words. Get the evidence.
getcited runs as a hosted MCP server. Your assistant calls it, getcited fetches the pages live, and the answer comes back as a report card: the score, every AI crawler's response, and each fix ranked by the points it recovers and the time it takes.
- 1
Add the connector
In Claude: Settings, Connectors, Add custom connector. In ChatGPT: Settings, Apps, developer mode. Cursor and VS Code take the URL in mcp.json.
- 2
Sign in with getcited
One click. No account, no password, no API key. getcited only reads the public pages you ask about.
- 3
Ask
“Audit my site for AI search.” “Who gets cited instead of me for this question?” “Write a passage that wins it, and prove it before I publish.”
Real results, fetched live by the hosted getcited MCP server.
AI visibility dashboards
Tell you that you lost.
$99 to $500 a month, weeks after the fact, with no idea which paragraph cost you the answer.
getcited
Tells you why, before you ship.
The passage that lost, the words it is missing, and a failing test in the pull request that would have cost you the citation.
Six pillars
From one question to a test suite.
01 · Citation Arena
Who does AI cite instead of you, and why?
Runs your page and your competitors’ through the retrieve-then-cite pipeline AI search uses: passage chunking, query fan-out, ranking and rank fusion. You get citation share, the exact passages the engine reads, and what your page is missing.
02 · Citation X-ray
Which sentences would AI actually quote?
Scores every sentence as quotable, weak or dead weight, with the reason: marketing fluff, starts with “It”, no specific fact, call to action, too long to lift.
03 · Citation tests in CI
Unit tests for AI search.
Write the questions your customers ask in getcited.yml with expectations like top3 or share>=30%. The GitHub Action fails the pull request when a change would cost you a citation.
04 · Regression monitoring
Know the day you lost the answer.
Every run is saved and compared with the last. The diff names the passage that dropped out, the competitor that took the top slot, the page that stopped loading.
05 · Truth
What is AI telling people about you?
Asks a model about your brand and checks every claim against your facts. Wrong claims become fixes you can publish.
06 · Audit
robots.txt says allowed. Your firewall says 403.
25 checks across reach, understanding, quotability and trust, including a firewall probe with real AI crawler user agents. Drafts the fixes: robots rules, JSON-LD, llms.txt.
One rewrite
0% → 80% citation share.
A marketing-heavy page against two competitors, then the same page rewritten from getcited's advice: answer first, the question's own words, real numbers.
Real-world study · 6 Nov 2026
5 questions. 27 ranking pages.
We ran getcited on the pages that rank for real questions, our own sites included. Every raw result is public.
100%
citation share for one focused page
On “How much does it cost to sell a structured settlement?”. Bankrate’s best passage ranked #36.
2 of 22
ranking pages turn AI crawlers away
robots.txt said allowed. Wikipedia’s GEO article refused ClaudeBot; the #1 page for “Should I block GPTBot?” refused GPTBot, ClaudeBot and PerplexityBot.
20%
quotable sentences on the median ranking page
Four in five sentences on pages that already rank are weak or dead weight to an answer engine.
Pricing
Free to run. Paid to have it run for you.
The tool stays open source. SCAIROS can run it for you every week, read the results, write the fixes and tell you the day you lose an answer.
Open source
Free
MIT licensed, run it yourself
- CLI: arena, X-ray, audit, truth, bench
- GitHub Action with job summary and badge
- Regression history in the Actions cache
- MCP server and agent skill
- Bring your own model key, or none
Citation Guard
$190/month
Run for you by SCAIROS
- 25 buyer questions tested every week
- Regression alerts: what changed and why
- Monthly live check of real AI engines
- Monthly review call
- Cancel any time
Citation Guard Pro
$490/month
Up to 3 sites
- 100 questions across 3 sites
- Two losing pages rewritten monthly
- Monthly brand truth check
- Slack alerts, priority support
Rather have it done for you? SCAIROS runs Answer Takeover sprints and white-label AI visibility for agencies.
Get getcited updates
New releases and study results, about once a month. Want it run for you instead? See AI Search Testing.
FAQ
Is the arena what ChatGPT actually does?+
It replicates the shape of the pipeline retrieval-augmented engines share: fan-out, passage retrieval, ranking, citation from a short context. It does not reproduce any single engine’s index or model. That is the point: it isolates the part you control, your passages against your competitors’ for the same question. getcited bench measures how well it predicts a real engine’s citations.
Do I need an API key?+
No. The arena, X-ray, audit and tests run offline and give the same result every time. A key only adds live answers from a real model, the truth check, and benchmarking against Perplexity.
How is this different from AI visibility dashboards?+
Dashboards sample AI answers and tell you, weeks later, that you lost. getcited tells you why at the passage level, before you ship, and fails the build when a change would cost you a citation.
Who builds it?+
Mohamed Said Yekhlef, AI systems engineer and founder of SCAIROS. SCAIROS also runs done-for-you GEO work if you would rather not do it yourself.