AI visibility platform
Make AIs recommend you
Your customers ask ChatGPT, Claude, Gemini, Perplexity, Mistral and Grok before they ask you. Epovest measures what the engines say on your prompts, shows you where to appear and what to fix, and proves what changed.
Views from the app. Your figures replace the sample at the first survey.
Real prompts, real answers
Your prompts are asked word for word through the engines' official APIs, and every full answer is kept.
Rates, not one-off screenshots
Each prompt is asked several times per check on higher resolutions: your citation rate rests on that many more draws.
Prepaid, no subscription
You buy credits and pay $0.10 per check. A paused tracker costs nothing and keeps its history.
Your customers ask an AI before they ask you
When someone asks an AI engine for a business like yours, a handful of names come back. You cannot shape what you do not measure. Epovest turns those answers into numbers you can track, and shows you how to move them.
The AI visibility platform
Getting cited by AI engines now has a name, several in fact: GEO (generative engine optimization), AEO (answer engine optimization), LLM SEO, AI visibility. Whatever you call it, it starts where SEO started: with a measurement. Epovest is the platform of the category: the numbers that tell you where you stand, and the levers to shape what the engines say.
The suite
Every step has its tool
The tools that carry the method, in the app and from your AI assistant alike.
Canon
Decide what the AIs should say about you, in one versioned text.
Tracking
Track what the AIs answer about you, at your cadence.
Surfaces
Register every place you speak, aligned with your canon.
Corroborations
Record the pages about you where someone else has the final say.
Atlas
The map of the sources the AIs lean on, ranked by priority.
Logbook
Every action dated, readable against your measures.
Quests
Across the whole methodWhat there is to do next, and where the work resumes.
The metric
AI Authority, the metric that says where to start
You know the sources the AIs lean on for your market. AI Authority tells you which one to work on first, engine by engine, on a scale everyone reads the same way.
The scale, in one picture
AA = 100 x detections of the source / detections of the engine leader
The source this engine leaned on most over the window. It sets the 100, and it is drawn from the entire corpus.
The engine leaned on it around six tenths as often as on its leader. Firmly in the part of the list worth working.
Cited regularly, well below the head. Often a specialist site, and often the easiest to be quoted in.
What do you want to track?
A tracker starts from your prompts, in your own words. The ways teams use it most; the prompts, the engines and the cadence stay in your hands.
Reputation
Know what the AIs say about you, and keep watch on it.
- One name to watch: a brand, a company, a person
- Answer rate and tone, check after check
- An email alert if a negative answer appears
Ranking
On your prompts, know who the AIs cite.
- A scoreboard of your keywords: you and the ones you compare
- Citation rate per engine, and share of voice
- New names suggested from the sources the AIs cite
Watch and archive
Follow what the AIs answer over time, and keep every answer.
- No keyword needed: your prompts are the watch
- Every full answer archived, with its sources
- A dated series, ready to re-analyse later
Each objective is the same configurator: the price always comes out of the grid, prompts × engines × checks × $0.10, plus $0.05 per analysed answer for each optional AI analyst.
Collection and reading are separate
A tracker archives raw answers. The analysts you graft on top (presence, share of voice, sentiment, custom analysis) read them without ever touching the record.
Graft analysts freely
Add or remove keywords and analysts whenever you like: your changes apply from the next check, and the record never moves.
The raw answers stay
Full answers and cited sources are archived check after check. Your series is never purged.
Structure is versioned
Changing the prompts, the engines, the cadence or the resolution starts a clean new version of the series: comparability is protected, the past is kept.
Integration
Drive Epovest from your AI assistant
Build and read your trackers without leaving the assistant you already use. One server address, two ways to connect depending on your tool.
Build a tracker
Describe what you want to measure: your assistant creates the tracker, with your prompts, your engines and your cadence.
Read your results
Citation rate, share of voice, full answers: your assistant pulls them and explains them to you.
Stay in control
Pause a tracker, adjust a keyword, prepare a top-up that you approve on the payment page.
Built for volume
Everything on the screen can be read and driven from your own tools.
As many trackers as you need
No cap. An agency runs one per client, and the only bound is your credit balance.
API and MCP, the same keys
The REST API v1 and the MCP server serve everything the app shows, in JSON, at 120 requests per minute per key.
Nothing is dropped
Every raw answer is archived and every series is kept, including while a tracker is paused.
Prepaid, no commitment
You pay for the surveys that run. No subscription, no tier to negotiate.
Pay per use, prepaid
No plans, no tiers, no subscription: a tracker costs exactly what it runs.
$0.10
per prompt, per engine, per check
Your credits pay for what runs. The suite itself, Canon, Surfaces, Corroborations, Atlas, Logbook and Quests, comes with the account.
Credits never expire · minimum top-up $10 · prices in USD
Our publications
What we publish on how the AI engines pick the names they recommend.
Guide
How AI labs choose the websites that enter their training corpora
Nobody at an AI lab hand-picks websites. A shared crawl, documented filters, deduplication, licensing deals and sampling weights decide. What each stage keeps, what it drops, and what that implies for the way you publish.
Study
When nobody writes about you, AI answers with your namesake's reputation
Twelve entities sorted by how much independent coverage exists about them, put to six AI engines on 6 August 2026 with a neutral question and an adversarial one: 144 answers. Every thickly covered entity drew criticism unprompted. On thinly covered names the criticism that comes back belongs to somebody else, sourced and stable.
Make AIs recommend you
Create a tracker in minutes: your prompts, your engines, your cadence. The first check runs the moment you start.
No subscription. Credits from $10 that never expire. Pause anytime: the history stays yours.