Use case · Watch and archive
What do the AIs answer on a topic, over time?
A market, a regulation, a destination, a technology: Epovest asks your prompts at your cadence, lines the answers up to show how the discourse moves, and keeps each one in full, dated, with the sources it cited. No keyword needed to start.
Views from the app. Your figures replace the sample at the first survey.
Answers are ephemeral, your archive is not
There is no public index of AI answers: once an answer changes, the previous one is gone. Epovest builds the record. Your prompts, asked at your cadence; every answer kept in full, with the sources the engines cite. When you need to look back, the series is there.
What you measure
Full answers
Every answer archived as the engine gave it, engine by engine, check after check.
Cited sources
The domains behind the answers, extracted at every check.
A dated series
The answers lined up over time: read how the discourse moves on your prompts.
Ready to re-analyse
The raw series stays complete and readable, ready to be reread whenever you need to look back.
The recipe
How to run an AI watch with Epovest
The full working guide for this use: topic prompts that let the answer move, a complete dated archive, and the Atlas as the map of who holds authority on the topic.
The prompts: a topic, not a brand
A watch asks the engines about a TOPIC whose answer can change: a market, a regulation, a technology, a destination. The ideal question admits several successive answers.
"What should a French SMB do about [regulation X]?"
"Where does [technology Y] stand: mature or not yet?"
"Is [sector Z] still worth investing in?"
"Top 5 [destinations for a team offsite]?"
Avoid closed factual questions whose answer will not move: a watch archives a discourse that evolves. List prompts, top 5s, rankings, are precious: their composition changes, and that is exactly what the series shows.
The settings that fit a watch
A watch optimizes for the archive first and the numbers later: the settings follow that order.
- Engines
- All of them. The point of a watch is triangulation: the same topic told differently by each engine, and the gap between their stories is information in itself.
- Cadence
- Set it to the speed of the topic: monthly for a regulation, weekly for a market, daily through the event, an election, a launch, a crisis. Come back down afterwards: the series continues in the tracker's new version.
- Resolution
- hd is enough as long as the deliverable is the archive of the discourse. Raise it the day you start counting, the appearance rate of a player, the frequency of an argument: a rate needs repetition to be stable.
- Analysts
- None is required to start: the tracker archives, you read. The lever specific to this use is Custom analysis: your instruction applied by an AI to every answer, list the risks mentioned, tell me the position taken on X. It is billed per analyzed answer.
The archive: what every check preserves
At every check, every answer enters the record in full, dated, with the sources the engine cites.
That is the raw material: AI answers have no public index, and when an answer changes, your archive is what keeps the previous one. Read the series as a thread: what appears, what disappears, recommendations that move, the new argument settling in. Filter by engine and by question; every quote in a report is checked against the very text of the answer, with its date.
What the archive makes possible
The record is a deliverable of its own: what an engine answered on a given day, in full, with the sources it cited, stays readable long after the answer has moved on.
Dated record
What an engine answered on 12 March, with the sources it cited, is read back in full. The date and the text are kept as they came, not a summary of them.
Dispute and competition
A false statement about your brand, or a ranking that leaves you out, is established at a precise date, with the complete text and the sources that fed it.
Disclosure and compliance
When you have to show what a recommendation rested on, the answer and its sources are already filed, engine by engine and check by check.
Diligence and tenders
Show what the assistants said about you over a period, drawn from a continuous series rather than one isolated screenshot.
Post-mortem
Put a dated action of yours next to a shift in what the engines say: the Logbook and the series share one record, so both ends read together.
Re-analysis
The answers are kept raw. A question you were not asking at collection time replays on the whole history, with no new check to run.
Each of these produces the same object: the full text of an answer, its date, its engine, and the sources cited alongside it.
The Atlas: the map of who holds authority on the topic
Who feeds the answers on your topic? The Atlas shows it, engine by engine, with the weight of each source.
Narrow the map to your watch project: the domains it lists are the topic's effective bibliography. Two readings belong to this use: the freshness sort, which shows entering sources, a domain that surges often announces a story that is changing; and the comparison between engines, the same question fed by different bibliographies. For a watch deliverable, this map is a result in itself.
Quantify along the way: keywords come later
A watch starts with no keyword. They arrive the day names emerge from the answers and deserve a numbered series.
A player keeps returning from answer to answer? Add it as a keyword: its presence series starts at the next check, and the archive already built does not move. Then switch discovery on: the recurring domains of the answers will be suggested, to accept or set aside. The qualitative watch becomes a measurement, without changing trackers.
Logbook and Quests: date the world, ritualize the reread
In this use, the Logbook does not date your actions: it dates the events of the world that can move the discourse.
The law passed, the report published, the competitor's launch, the incident: recorded at their date, they sit on the series and rereading becomes a reading of causes. In Quests, pose your rituals: reread the series after [the deadline], add [the emerging player] as a keyword, produce the monthly digest. The file keeps the watch's rhythm for you.
Take the material out: the archive is also read by machines
A watch deliverable rarely ends in the app: it ends in a digest, a report, a table.
The raw series reads through the API and through MCP, answer by answer, sources included: plug your AI assistant into it and ask for the month's digest, the gaps between engines, the positions that moved. The archive stays yours, complete, readable months later with the questions you were not yet asking at collection time.
Worked example
A regulation evolves, the discourse follows
Six months of monthly watch on [regulation X], then the deadline nears: switch the cadence to weekly, date the deadline in the Logbook. The answers change tone, two engines cite a new domain: the Atlas brings it up in the freshness sort, you add it as a keyword and its series starts. At the deadline, Custom analysis summarizes the position of every answer, and your digest quotes dated texts, not memories.
With your AI assistant
Your AI assistant runs this recipe with you
Connect Epovest to the assistant you already use, one-click connector or API key. It creates trackers, reads the results and keeps the registers with you; a payment only ever happens with your approval.
The instruction to give it
Connect to Epovest and set up an AI watch on [my topic]: four prompts whose answer can move, all engines, monthly in hd, no keyword. At every check, summarize what changed and which sources are entering.
$0.10
per prompt, per engine, per check
Pricing
Whatever you configure, the unit does not change. Your total shows live in the configurator before anything runs; you start whenever you want, from a $10 top-up. No subscription: only the checks that run are debited.
Start light, densify later
Begin with a few prompts and one or two engines, then raise the cadence, the resolution or the number of engines when the topic heats up. The series already collected is kept.
The other use cases
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 do not expire. Pause anytime: the history stays yours.