Sentl.AI
Sign up
LLM reputation management

Monitor what the AI models say about your brand, and build your reputation strategy on it.

Sentl.AI puts your questions to the models — for your own brand, or for the ones you look after — grades each answer 1 to 5 against your rubric, and shows which sources they read to get there.

Sign up See how a run works

€299/month for one brand, €1.799 for an uncapped portfolio.

Example project — illustrative numbers, not a customer result

Northwind Coffee run of 28 July · 4 models · 20 answers
Completed

Reputation score

3.8/5

Band Fair +0.4

Score from 0 to 5 for each dimension, run of 28 July
Reputation4.2
Reliability3.6
Positioning3.1
Service4.5
Price2.8
The problem

Checking by hand works once and stops there.

Four models, the same prompts pasted by hand, screenshots into a slide. It holds for one afternoon: no two rounds are comparable, and nothing about it repeats next month — or on the brand next door.

  • Someone asks what the models say — a client, a board, a colleague — and the honest answer is "we're looking into it". More ask every quarter.
  • It is manual end to end, so nobody else can run it and come back with the same thing.
  • Nothing shows whether last quarter's work moved anything, or which outlet to work on next, so it gets defended on trust.
Four models, one afternoon, every time
What you monitor

You choose what gets measured, and what a good answer is.

One project per brand. You name the dimensions, write the questions, and write the rubric the judge applies to every answer — the way the people who test AI features for a living have graded model output for years: what a 1 looks like, what a 5 looks like, and a separate model deciding.

  1. 01

    A project per brand

    One brand, the models you want queried, how often the run repeats, and how many times each question is asked.

  2. 02

    Dimensions you name

    The areas you already report on. Each weighs the same in the final score.

  3. 03

    Questions you write

    The questions this brand's buyers actually put to a model, grouped by the moment they belong to: no idea yet, comparing, checking a rule, resolving a doubt.

  4. 04

    The rubric you write

    Applied unchanged to every answer, from every model, on every run.

Questions you might write

Written the way a buyer would type them. The brackets are where the name goes.

  • Where you stand in the categoryWhich coffee suppliers would you recommend for a Milan office?
  • What the models think you doWhat does [BRAND] do, and who is it for?
  • How they describe youHow would you describe [BRAND] in a few lines?
  • Against a named competitorBetween [BRAND] and [COMPETITOR], which would you pick for a 200-desk office?
  • What they warn aboutIs [BRAND] reliable? Any controversies or negative reviews?
  • Which sources they lean onWhich third-party sources talk about [BRAND]?

What a rubric looks like

Yours will say something else. That is the point of writing it.

Example rubric for one question
ScoreThe answer…
1advises against the brand, or criticises it
2does not mention it at all
3mentions it neutrally, among others
4mentions it positively
5makes it the main recommendation

A 2 and a 1 are different problems. Silence is filled with content the models can cite. An answer that argues against the brand has to be corrected where it comes from.

"Another AI grading an AI." Never the model that answered. It reads one answer at a time, does not browse, and writes its reasoning beside the score.

Results

What comes back from a run.

Where the models treat the brand well, whether the score is moving, and behind every number the answer that produced it.

Example project — illustrative numbers, not a customer result

Score by dimension · run of 28 July
The radar above, read exactly. Score from 0 to 5.
DimensionScoreBand
Reputation4.2Strong
Reliability3.6Fair
Positioning3.1Fair
Service4.5Strong
Price2.8Weak
Trend · last six runs
  • Reputation
  • Reliability
The same series, read exactly. Full 0–5 axis, not truncated.
DimensionFebMarAprMayJunJul
Reputation3.13.33.23.63.94.2
Reliability2.82.93.23.13.43.6

"The models answer differently every time." They do — which is why each question is asked more than once in a run, and how much the answer moves is itself a measurement. The questions and the rubric are held still while it does.

Tone is read on each answer. There is no project-level sentiment percentage: that number would be comfortable and would mean nothing.

Sources

Which sources the models read, and what that tells you to do.

The score says where the brand stands. The sources say why, and which outlets are worth working on — which is what turns a measurement into a digital PR or GEO brief.

Example project — illustrative numbers, not a customer result

What one model cited · one answer on Reliability
The sources under a single graded answer
SourceCited for
industryweekly.examplethe 2024 recall
northwindcoffee.examplethe product range
reviewsite.exampledelivery complaints
What recurs across models · how many of the four cited it
The same outlets, by dimension. A source three models lean on is where the work goes.
SourceRep.Rel.Pos.Price
industryweekly.example341—
northwindcoffee.example2132
reviewsite.example13—2
coffeeguide.example——31

A cited source is not a supporting one: the models sometimes name a page that does not say what the answer claims. That is why the answer, the citation and the judge's reasoning stay together.

Editorial content feeds 61% of AI answers about the world's 100 most valuable brands, against 44% for their own sites: the models read the sources PR has always worked on.

"We already do media monitoring." That tells you what was published. This tells you what the models took from it, and which placements never reached an answer.

And then you know what to work on

  • Fill the silence. Where the models do not mention the brand, give them something of its own to cite.
  • Correct the lean at the source. Where the models start out against the brand, the work is on the outlets feeding that view, before it sets.
  • Move the advantage upstream. Where the brand already wins, but only when asked directly, the argument belongs in the earlier questions.

The third is the common case: in one study a brand scored 3.7–4.3 where it was named and 2.2–3.1 where the buyer had not decided yet — which is where the choice is made.

How this differs

Counting mentions leaves out what the model actually said about the brand.

The tools you have already looked at report the four numbers on the left, on definitions that are theirs and identical for everyone using them.

The tools you have already looked at

  • Visibility — how often the brand appears
  • Position — where it ranks when it appears
  • Sentiment — on the vendor's scale
  • Share of voice — against competitors

Sentl.AI

  • A written rubric: what earns a 1, what earns a 5
  • A separate model grading, one answer at a time
  • Reasoning written before the score
  • The criteria kept with the result they produced
Setup

Set a brand up by talking to your assistant.

Connect Sentl.AI to your assistant over MCP and the setup happens in that conversation — project, dimensions, questions, rubric — in minutes rather than an afternoon. When a run finishes, the same assistant writes it up.

Which assistants. The ones that speak MCP: Claude, ChatGPT, Gemini, Perplexity, Grok.

  • Nine tools, and three of them write

    Reads projects, runs and results; with the right permissions, creates one and starts a run.
  • OAuth, or a token you paste

    Log in and consent, or paste a personal access token.
  • Scoped to one organization

    One organization per connection, permissions recomputed from your role on every request.
Who is running it

Food, automotive, fashion and multi-utility brands are running it, and so are agencies.

None of them can be named yet, which is why this page carries no logo wall. An anonymous quote would be worth less than saying plainly who is using it.

Where to start

Run it on your own brand first. What comes back is what you would put in front of a board, or a client.

Pricing and credits

Three plans. What changes is how much you can measure.

Save up to 17% with an annual commitment

The plan sets the credit allowance and how many projects stay active. Everything else is in all three.

Brand

One brand.

€247 /month

Saving €624 a year

  • 4.200 credits a month
  • up to 3 active projects
  • about 6 full runs a month
Start with Brand

Team

Several brands or markets.

€497 /month

Saving €1.224 a year

  • 9.200 credits a month
  • up to 8 active projects
  • about 13 full runs a month
Start with Team

Agency

For a roster

An uncapped portfolio of brands.

€1.497 /month

Saving €3.624 a year

  • 30.400 credits a month
  • unlimited active projects
  • about 44 full runs a month
Start with Agency

A full run is four models answering ten questions, judge included: 680 credits.

In every plan
  • All AI models
  • MCP connection
  • API key
  • Team members
  • Credit history
  • Scheduled runs
  • Credits renew, they do not accumulate. Extra packs need no plan change.
  • Changing plan. Moving up is immediate, moving down starts at the next renewal.
  • One subscription per organization. Most teams keep every brand inside one.

Questions we get asked

Can I not just ask ChatGPT myself?

You can, and you get one answer, from one model, on one day. What this adds is the same questions put to every model on a schedule, graded against criteria you wrote, with a history that shows whether the work moved anything.

It costs more than the tools we have already looked at.

It does. The Agency plan is €1.799/month for an uncapped portfolio — about 44 runs a month, roughly €180 per brand across ten brands and €90 across twenty. The Brand plan is €299/month for a single one.

Where does the data live?

Everything runs on European servers. The models you query are third-party services running their own inference wherever they do — that part is theirs, not ours.

Can we resell this to our clients?

Yes, under the reseller partner plan — terms depend on your roster, so they are agreed rather than published. The product itself stays Sentl.AI-branded; what reaches your client is the deliverable you produce. Write to reseller@sentl.ai.

How long does it take to set up the first brand?

Minutes through your assistant: connect Sentl.AI over MCP and it walks you through project, dimensions, questions and rubric. The rubric takes thought rather than time, and you write it once per brand.

What happens if we change the rubric halfway through the year?

From the next run. Every score keeps the rubric that produced it, so the history is not rewritten by a criterion you refined in June.

Which models does it query?

The ones you pick from the catalogue — GPT, Claude, Gemini, Llama, Mistral, Grok and Perplexity among them. Change the selection later and completed runs stay as they are, so the comparison holds.

Set up one brand and read the first score.

One dimension, three questions, a rubric. Start the run and read what four models said.

Set up your first brand

Set up in minutes · European servers