LLM brand visibility: retrieval is the part you can fix

A model knows two things about your brand: what it absorbed during training and what it fetched a second ago. You cannot change the first. The second is served by your own web server.

Scan a URL freeNo account, no card. You get the score and the fix list.

Your business

Products, services and policies

What is missing

Details a customer needs

A clearer answer

Information you can publish

Services page, bytes an agent reads
89.6%
docs/demo/agent-before-after-document.md (13,581 → 1,412 bytes)
Tokens saved on the same page
9.7×
docs/demo/agent-before-after-document.md

Training and retrieval are different problems

Brand teams often treat an assistant’s wrong answer as one problem. It is two, with different time constants and different remedies, and the useful diagnostic is whether the answer changes when the model is allowed to browse.

Training knowledgeRetrieved knowledge
Where it came fromA corpus frozen at some past dateA live fetch of a URL
How stale it can beArbitrarilyAs stale as your page
Can you correct itNot directlyYes, by changing the page
How long a fix takesA model generation, if everThe next fetch
Who controls itThe model providerYou

This is why the practical brand strategy is to make retrieval win. A model that can cheaply fetch a current, unambiguous page about you has less reason to fall back on whatever it half-remembers, and that is the entire mechanism available to you.

Making the retrieved version authoritative

Three properties decide whether your page is used once it is fetched, and all three are ordinary engineering rather than content strategy.

  • Reachable. No bot rule, CDN filter or challenge page between the agent and the content. Most blocks on assistants were inherited from an old anti-scraping policy rather than chosen.
  • Complete. The facts a person would ask about are present in the text a parser sees, not rendered client-side and not implied by an image.
  • Cheap. On the reference document capture, the same page fell from 13,581 bytes to 1,412 and the first fact moved from character 10,152 to line 8. Cheap pages survive truncation; expensive ones get summarised badly.

What to report internally

Report the measurements, label the samples, and resist converting the two into one number. A single blended "visibility score" combining exact server data with sampled prompt results is the metric most likely to be quietly wrong in a quarterly review.

  1. Baseline the templates. One readability score per page kind, taken before any change.
  2. Report fetches and answer rate monthly. Exact, from your own logs, and defensible.
  3. Track the miss log as a backlog, not as a metric. It is a to-do list that happens to be ranked by demand.
  4. If you sample assistant answers, say so on the slide. It is indicative, and treating it as a target invites optimising for a number nobody controls.

What this does not do

Frequently asked questions

What is LLM brand visibility?

How well large language models can find and correctly state facts about your brand. It splits into training knowledge, which is frozen and not correctable by you, and retrieved knowledge, which comes from a live fetch of your pages and is entirely yours to fix.

Why does an assistant describe our brand with outdated information?

Usually because it fell back on training knowledge instead of fetching you, and the common reason for that is that fetching you did not work or did not pay. Check whether the answer changes when the model is allowed to browse: if it does, the fix is on your server.

How long until a change to our site shows up?

The retrieved half changes as soon as the page is fetched again, which depends on the platform and is outside your control. The training half does not change on any timeline you can plan around, so do not budget against it.

Is there anything we can do about training data?

Very little, honestly. Publishing clear, current, machine-readable facts is the only durable input, and its effect is on future corpora rather than on the model answering today. Anyone offering to edit what a shipped model already believes is selling something that does not exist.