More and more buyers ask ChatGPT, Gemini, Claude and Perplexity what to buy, or send an agent to buy for them. The answer names a handful of businesses. EarnCitations makes sure you're one of them — and only charges when the citation lands.
new here? how to use it, in order · or use it from your agent: MCP server, CLI & API
Every AI answer is assembled from two things: what the model looks up and what it already believes. Most tools watch the first. We work on both — on your site, and across the web that AI actually reads.
Review sites, Reddit threads, listicles, docs, directories, podcasts, your rivals' comparison pages. Winning means earning a place in those sources — honestly.
Your old pricing, a founder who left, the city you moved from. We separate what each model recalls from what it retrieves, then correct it at the source.
Agents already pick vendors, read docs, compare plans and check out. If an agent can't understand, trust and buy from you, it buys from someone it can.
Start with the plumbing. Our audit checks 26 AI crawlers against your robots.txt, reads your pages the way most AI crawlers do (the server HTML, no JavaScript), reads your schema, llms.txt and Fact Ledger, and hands you a ranked fix list.
Also free: the AI crawler directory (26 agents, sourced to each operator's docs) with a robots.txt tester and builder, an AI log analyzer (which agents really came, fake bots, and visitors who clicked through from ChatGPT or Perplexity), a question map (the buyer questions your site should answer, each checked), a question doctor (why AI doesn't cite you for one buyer question, and the fix), what AI sees on any page, a delivery check (did the fix you shipped reach AI crawlers?), an answer check (is what AI says about you backed by its sources?), an answer log (how often answers name, recommend and cite you, over time), an AI price check (what price an assistant will read for you), a head-to-head against a competitor, an llms.txt generator, an Organization schema generator, the state of AI-readiness in 124 software companies, and 150 independent hotels and restaurants.
From our research this week: only 27 of 116 software sites say plainly whether there's a free trial (buyer questions), and 13 of the 34 Markdown copies of pricing pages that sites hand AI agents leave out prices the page shows people (Markdown copies).
Trackers tell you where you stand. Agents write you a blog post. EarnCitations is the whole machine — from a simulated market of buyers to an invoice that only counts citations we can prove. We're pre-launch, so each system is marked: live, open-source prototype, or in development.
Thousands of simulated buyers — persona, intent, budget, locale, device, memory — holding real multi-turn conversations with every engine. Not "50 prompts". Your market.
Ask each model with retrieval off, then on. What it remembers versus what it looks up, scored claim by claim, with drift alerts on every model release.
For every topic and engine: the domains that get cited, the upstream pages that feed them, and the exact holes where your name should be.
Shadow retrieval replicas predict the lift of a change before you ship it, with an honest confidence interval. Synthetic-control attribution after.
Pages, edits, schema, llms.txt, docs, Google Business Profile. Shipped as a pull request, a CMS draft or a listing update. Nothing goes live without your merge.
Original data studies, expert commentary, podcast and journalist pitches, disclosed community answers, review asks to real customers. All the places AI reads that aren't your website.
A signed, versioned /.well-known/facts.json of what's true about your business — with evidence links agents can verify. Open spec, free generator.
Make your product chosen, understood, used and bought by agents: MCP endpoint, machine-readable pricing, agent checkout, setup-from-docs tests.
A calendar of crawl and training cutoffs. Make sure your canonical facts live in the high-weight corpora before the next model generation freezes.
The Answer Index: a public, reproducible benchmark of who AI recommends in every category, re-run monthly. Methodology and harness open-source.
Every change runs the same seven steps. The last one is the one we're building toward: bill only for what attribution says we caused.
Your buyers ask every engine, in conversation.
Missing page, weak page, wrong belief, or a source you're absent from?
Predicted lift and confidence for every candidate fix.
PR, CMS draft or listing. You merge.
Studies, pitches, reviews, community — disclosed.
Synthetic control against untouched questions.
Only for verified net‑new citations.
Working today, free: find why you're not the answer with the question doctor, ship the drafted fix, confirm crawlers got it with the delivery check, read what assistants say now with the answer check, and follow it over time in the answer log (the method, step by step). Selling to developers? Check a coding agent could set you up from your docs with the docs check. Forecasts, pull requests, outreach and the synthetic control are still being built: today the fixes come as files you copy in.
Before Forge writes a word, Oracle replays the change against shadow replicas of each engine's retrieval and tells you what it's worth. After it ships, we measure against the questions we didn't touch — so the lift is ours, not the weather's.
The AEO market grew up fast: first visibility dashboards, then agents that ship content as pull requests. Here's where EarnCitations goes past both.
| Visibility dashboards | AEO agents | EarnCitations | |
|---|---|---|---|
| What gets measured | Single prompts | Single prompts | Multi-turn synthetic buyers, voice and agents |
| Knows why you lose | — | Reads cited pages | Citation supply chain + weights‑vs‑web belief map |
| Predicts lift before shipping | — | — | Forecast with confidence interval |
| Ships fixes | — | GitHub PRs | PRs, WordPress, Shopify, Webflow, Google Business Profile |
| Works off your site | — | Outreach drafts | Earned-media desk: studies, press, reviews, community |
| Open standard for facts | — | — | Fact Ledger, signed & open |
| Agents as customers | — | On the roadmap | Chosen → understood → used → bought |
| Next model generation | — | — | Epoch training-window planning |
| You pay for | Prompts tracked | Prompts + work credits | Citations earned |
The EarnCitations column is the full product. What exists today is marked on each module in §3: the audit, fix kit, question doctor and map, what AI sees, answer check, delivery check, watch, log analyzer, Fact Ledger, WordPress plugin and the open-source prototypes.
Engines and platforms work to catch manipulation, and a brand caught cheating can lose more than it gained. So we hold ourselves to four rules, and we publish the receipts.
No prompt injection, no text only bots can see, no cloaking. What AI reads is what humans read.
No fake reviews, sockpuppets or undisclosed posts. Community answers are signed by real people, disclosed.
Every sentence Forge writes links to evidence a human can check. No source, no claim.
We follow Wikipedia's conflict-of-interest rules and community norms. We fix facts through the front door.
Publish your facts once, signed, in a format any agent can verify. And run all of EarnCitations from the agent you already use.
// Fact Ledger v0.1 — open spec at earncitations.com/ledger { "subject": { "name": "Northwind CRM", "url": "https://northwind.example" }, "facts": [ { "id": "pricing.starter", "claim": "Starter plan costs $29 per user per month", "value": { "amount": 29, "currency": "USD", "per": "user/month" }, "evidence": ["https://northwind.example/pricing"], "validFrom": "2026-09-01" }, { "id": "integrations.gmail", "claim": "Two-way Gmail sync on every plan", "evidence": ["https://northwind.example/docs/gmail"] } ], "signature": { "alg": "Ed25519", "keyUrl": "/.well-known/facts.pub", "sig": "…" } }
# 1. audit any site — free, open source (npm package coming; today run from the repo) node bin/earncitations.mjs audit northwind.example # 2. draft the fixes: llms.txt, robots.txt AI rules, JSON-LD, facts.json node bin/earncitations.mjs fix northwind.example # 3. hand it to your agent claude mcp add earncitations -- node /path/to/bin/earncitations.mjs mcp # then just ask: > audit northwind.example and draft the fixes for everything that's failing
A public benchmark of AI recommendations, category by category, with open methodology and an open harness. Below is a preview of the format with fictional brands; the first real edition is in the works.
The audit and fix kit are free: they show why AI doesn't name you, and you fix it yourself. A paid plan checks your buyer questions every week, does the fixing and shows you it worked. Paid plans open soon; early access is open now.
Free: find out why AI doesn't name you, and fix it yourself.
Paid: we check every week, do the fixing, and show you it worked.
Every week, a paid plan gives you
Starter. Checked weekly on every major AI engine, about 220 question checks a month. 14-day free trial when paid plans open, then month-to-month.
Get early accessPaid plans open soon. Ask for early access, say which plan you want, and we'll set you up first. The free tools work today.
Enterprise. More than 1,000 questions or custom terms? We'll build a plan around you, with a citation guarantee.
Talk to us| Plan | Price | Buyer questions | Checked | Includes |
|---|---|---|---|---|
| Audit | $0, forever | 25 | When you run it | Unlimited AI-readiness audits, Fact Ledger generator, listed in the Answer Index |
| Starter | $99 a month | 50 | Weekly, every major engine | The full Citation Stack, unlimited seats |
| Growth | $299 a month | 400 | Every other day | Everything in Starter, plus API, MCP and webhooks |
| Scale | $699 a month | 1,000 | Daily | Everything in Growth, and your agents run the loop |
Paid plans open soon, with a 14-day free trial, then month-to-month; ask for early access. Prefer to pay only for results? Earned has no base fee: $2–$60 per verified net-new citation, with a monthly cap you set. See full pricing.
An AI answer that names your business or links to a source about you when responding to a real buyer question. We count it per engine, per question, across repeated samples — answers are non-deterministic, so one lucky run doesn't count.
Attribution. For each change we ship or earn, we hold back comparable questions as a synthetic control, sample every engine repeatedly, and only bill citations that (a) persist across samples, (b) trace to a page or source we shipped or earned, and (c) exceed what the control predicts. If we can't prove it, it's free.
Yes. Opening hours, locations, prices and services are the facts that go stale fastest, and they're exactly what the Fact Ledger states with a source and a date. Today the free audit, the LocalBusiness schema generator and a WordPress plugin work for any business; Shopify, Webflow and Google Business Profile publishing are in development.
Never. Every page, edit, pitch and community answer waits for your approval. On-site changes arrive as a pull request or a CMS draft.
Some of the plumbing overlaps — crawlability, structured data, good pages. But AI answers name three businesses instead of listing ten links, pull heavily from third-party sources, and mix retrieval with model memory. Ranking #1 on Google doesn't guarantee a citation. We measure and work on the AI side directly.
Lighthouse's Agentic Browsing category (experimental, Chrome 150 or later) loads your page in a real browser and checks what an agent operating it needs: WebMCP tools, an accessibility tree with names and labels, a stable layout, llms.txt and an ard.json manifest. Our audit reads what AI crawlers that don't run JavaScript get: which AI agents robots.txt lets in, your structured data and prices, and whether your pages answer buyer questions. It checks llms.txt, ard.json and the WebMCP attributes on your forms too, from the HTML. The two cover different ground, so run both.
That's the point. Everything is an MCP tool and a typed API. Your agent can run audits, read the plan, ask for forecasts and open the PRs.