Your AI is fast.
We make its answers bankable.
The numbers superpower for the AI you already use
60 seconds to connect · minutes to answer · as-filed
Works inside Claude, ChatGPT & Copilot · Prefer paper? Get a sample report →
Fast. Trusted. Affordable.
Pick two — that was the rule.
The terminal seat
Great data — priced for a trading floor, not a deal team.
Pulling EDGAR by hand
Right numbers, right price — but analyst-days per comp set.
The raw chatbot
Answers in seconds — invented peers, no audit trail.
What we deliberately don't do
No estimates, no forecasts, no valuation opinions — benchmarking over as-filed facts. Comparability limits are named per metric. Every stat here comes from the live registry.
Coming soon Trading multiples — EV/EBITDA, P/E and EV/Revenue as full-population percentiles, computed on licensed end-of-day market data with the same per-figure evidence chain.
Professional formatting proves nothing.
It even invents the sources.
On Aug 27, 2026 we asked a leading AI assistant for two benchmark deliverables — a peer spreadsheet and a cross-source ranking — then verified every single citation against SEC EDGAR and USAspending. The output looked investment-committee-ready. The evidence chain was fiction.
Finding 1 · The lineage was theater
The spreadsheet’s own header: “Every metric mapped directly to primary SEC Form 10-K accession numbers.” We checked all 26 against SEC EDGAR. One mapped. The methodology sheet described a bulk-data extraction that demonstrably never happened.
Finding 2 · Real and invented, side by side
In the same table: one company’s revenue correct to the million — the next one’s matching no fiscal year on record. A “ranking” that isn’t even sorted by its own ratio column. Same font, same confident tone. A reader cannot tell the copied row from the invented row.
Finding 3 · What it got right is the trap
The one number a government website publishes as a finished figure, it copied to the dollar. LLMs are excellent at copying published numbers — and produce fiction wherever a number has to be computed: populations, ratios across sources, rankings. Exactly the numbers a deal team needs.
This is why peerset makes AI answers bankable: if the proof isn’t attached, you are the fact-checker. Every peerset figure ships with its SEC accession and a live recompute link — like this percentile table over all 218 US software filers, the population math no assistant can copy from anywhere. Reproduce the test yourself: ask your AI for a peer table, then click its sources.
One plugin. Everywhere your team works.
peerset.ai is not another chatbot — it is a numbers superpower for the AI you already use: citable answers your team can rely on. Connect it once and it lives inside Claude, ChatGPT, or Copilot — everyone at the firm can ask, nothing new to learn.
One connection, whole team
No logins to manage, no seats to count.
Open standard
Switch AI tools tomorrow — peerset.ai comes along.
60-second setup
One link, pasted once.
What you get for your money — concretely.
Cited answers
“Your 12% net margin ranks at the 74th percentile among 92 US semiconductor makers” — with a footnote per figure, down to the SEC accession number.
Board-ready XLSX
Every answer exports to Excel: Summary, Peers, and a Sources sheet where each number carries its filing link. Download a real sample →
Recompute, live
Computed figures (percentiles, margins, growth) link to the exact public query — result, inputs and SQL. See one →
For your IT team
peerset.ai is an MCP server (the open standard for connecting data to AI assistants). Add this endpoint in your assistant's connector settings:
https://factanker.com/mcp
No API key during preview. Rate limits apply.
One day. Full access. Free.
We are in open beta — full access, honest rough edges, founders on the other end. The full peerset.ai chat, on our site — every peer unlocked, every citation live. Your email is your username. No card, no sales call.
We email your access link. Your day starts when you first sign in — not when you request it.
Need it on paper? Run a sample benchmark report.
The same engine, on paper: a 4–6 page PDF — percentiles, peer tables, every figure cited.
Built on a public fact registry. Check us before you trust us.
peerset.ai runs on its own public registry of as-filed facts. Every number resolves to a permanent URL.
Per firm. Not per seat.
Every plan: unlimited questions, whole team, all citations. A terminal is $30k+ per seat, per year.
- Full access, all peers unlocked
- Expense it — no procurement
- Everything in the day pass
- Plugin for Claude, ChatGPT & Copilot
- Saved peer sets, quarterly refresh
- Everything in monthly
- Restatement change-feed
- Priority coverage requests
Start with the free day — upgrade in one click. Enterprise (own tenant, EU/UK universes, SSO): talk to us.
Talk to the people who built the registry.
One email — answered by the founding team, not a funnel.