Know what to fix

Can AI assistants look up your brand's real facts, before a wrong answer costs you deals?

Wrong AI answers about your price, features or integrations cost you demos, renewals and support time. We publish your product and page facts as a small server AI assistants can ask, so one that looks you up gets your current facts. You can switch it off any time.

AI Agents settings showing the agent switched on, MCP connector address, and sample Q&A preview of what AI assistants can look up
One switch lets AI agents look you up. Your data, read-only.
  1. 1One switch publishes it, and switches it off again
  2. 2The address you share, and add to your llms.txt
  3. 3A preview of what AI assistants can look up

What you do with it

  • Try each lookup inside the app and see exactly what an AI assistant would get back
  • Switch it on to get a public address for your brand facts
  • Paste the address into your llms.txt so AI tools can find it

What it can't do

  • AI assistants don't find MCP servers on their own yet. There is no standard for that today.
  • Listing it in your llms.txt is the practical way to be found right now. A descriptor on your own domain is planned.
  • It answers from what you already publish. It can't add facts you haven't written down.
  • It can't make an assistant use your facts or name you.

What a wrong AI answer about your brand costs: three costs you can measure

When an AI assistant gets your price, a feature or your integrations wrong, the cost shows up in three places. Each one can be worked out from numbers you already have.

Lost pipeline. Buyers who read the wrong answer and never ask for a demo. Work it out as: buyers who saw the wrong answer × the share who would have asked for a demo × your demo-to-deal rate × your average contract value. The first number comes from the questions you track: how often the wrong answer appears, and how many AI visits those questions bring.

Churn risk. Customers who ask an assistant about a feature at renewal and are told you don't have it, or that it costs more. Work it out as: accounts up for renewal that rely on the misdescribed feature × their renewal value × the extra chance they leave. Tag churn and downgrade reasons in your CRM so the last number is a count, not a guess.

Support load. Tickets that start with "ChatGPT told me…". Work it out as: tickets a month that quote a wrong AI answer × handling time × hourly support cost, plus escalations × escalation cost. Add a ticket tag for it today; you need a baseline before you fix anything.

A worked example, with made-up numbers

These inputs are invented to show the arithmetic. They are not results from a customer.

Say assistants tell buyers your product has no single sign-on, when it does. 300 buyers a month see that answer. 3% of them would have asked for a demo, 20% of demos become deals, and a deal is worth €12,000 a year. That is 300 × 3% × 20% × €12,000 = €21,600 of pipeline a month.

Say 20 accounts up for renewal this quarter need single sign-on, each worth €12,000, and the wrong answer adds a 5% chance that each one leaves. That is 20 × €12,000 × 5% = €12,000 at risk this quarter.

Say 40 tickets a month quote the wrong answer, each takes 15 minutes, and support costs €40 an hour. That is 40 × 0.25 × €40 = €400 a month, before escalations.

Put your own numbers in. The point is to know which wrong fact is worth fixing first, not to produce one big figure.

What the server changes, and how to check it worked

The server does not rewrite what an assistant says. It gives an assistant that looks you up a direct, current source to answer from. If an agent asks "does it support single sign-on?", search_brand_knowledge returns the matching lines from your own product facts, and get_page returns your security page as clean text with its address. An answer built from that can quote you and link to you.

To check it worked, track the exact question that was answered wrongly, and look for four things in later answers: the fact is now right, the answer cites one of your pages, it stays right across the assistants you track, and it stays right on repeat runs, not just once. Keep the before and after answers side by side in your report.

Then go back to the three costs above. Compare demo requests from AI visits, the tagged churn reasons and the tagged tickets for the months before and after. Other things change at the same time, such as launches, pricing and seasons, so treat the result as a sign, not proof.

What your brand MCP server shares

Five lookups. Each one can read, never change. An AI assistant can search your product and marketing facts. It can list the pages you've made available. It can fetch one page as clean text. It can read your llms.txt. And it can read the structured data for a page or the whole site.

The tool names are search_brand_knowledge, list_pages, get_page, get_llms_txt and get_schemas.

Nothing from your Targetlytics account is shared. No analytics, no competitor data, no questions you track. It is the same material behind your llms.txt, schemas and public pages.

How an MCP server differs from llms.txt

llms.txt is a file. An AI assistant has to know it exists, download all of it, and find the part that matters.

An MCP (Model Context Protocol) server is something it can ask. The assistant asks one question and gets one scoped answer. Both serve the same facts, so they stay in step.

This is not the same as the Targetlytics MCP connector. The connector brings your Targetlytics data into your own assistant. The brand MCP server shares your public brand facts with everyone else's assistants.

Frequently Asked Questions

Questions about letting AI assistants look up your brand

A Model Context Protocol server for your brand. It lets an AI assistant look up your products, pages and facts with a direct question, instead of reading your site.

Split it into three costs. Lost pipeline: buyers who saw the wrong answer × demo rate × deal rate × contract value. Churn risk: renewing accounts that rely on the misdescribed feature × renewal value × extra chance they leave. Support load: tickets that quote a wrong AI answer × handling time × hourly cost. Track the wrong answer, tag tickets and churn reasons, and compare before and after the fix.

Not yet. There is no discovery standard today. Listing the address in your llms.txt is the practical signal for now, and we write that line for you.

It can read the brand facts you already publish, and nothing else. It shares nothing from your Targetlytics account: no analytics, no competitor data.

Yes. Every lookup runs inside the app, behind your login. You see exactly what an assistant would get before you make it public.

Yes. Switch it off and the address stops answering. Nothing is built into your site.

Let AI assistants ask you directly

Start with a free audit of how easily AI can read your site today.

Get a free audit

Free AI visibility audit

Find out what ChatGPT tells your buyers about you.

Where you're named, what AI gets wrong, and the first three things to fix. Emailed within two hours.

  • Within 2 hours
  • No credit card needed
  • Free
Prefer to talk? Book a 30-minute call
  1. Your site
  2. Your market

We'll send your report to this address