EU AI Act transparency duties apply now; high-risk duties from December 2027. Check your exposure
Insights About us Careers
Contact us
AI for Marketing and Growth

AI Visibility and Brand Monitoring in LLMs

Knowing what AI assistants say about you when a buyer asks — whether you appear at all, how you are positioned against competitors, and whether what is said is even true.

4 to 8 weeks
Setup
Retainer
Commercial model
Accuracy
Tracked separately

The uncomfortable version of this problem is not absence. It is an assistant confidently telling a prospect that you do not support a feature you shipped last year, or that your pricing starts at a number you have never charged, and the prospect believing it because it arrived without a source to doubt.

In one paragraph

AI visibility monitoring tracks how large language model surfaces respond to questions relevant to your category: whether your organisation is mentioned, in what position relative to competitors, with what characterisation, on the basis of which cited sources, and whether the claims made are accurate.

Four things worth tracking, in order of urgency

QuestionWhy it mattersHow we measure
Is what they say about us true?Inaccuracy costs deals directlyClaim-level checking against your own facts
Do we appear at all?Absence from the consideration setPresence rate across a tracked prompt set
How do we compare?Position against named competitorsShare of mention and characterisation
What sources are cited?Shows where influence actually comes fromCited-source analysis per prompt

Most vendors lead with the second row because it makes the nicest chart. The first row is the one that loses revenue this quarter, so it is where we start.

Building monitoring that means something

Prompts from real buyers, not from a keyword tool

Assistant prompts are long, conversational and often comparative or conditional — 'what's the best option for a mid-size manufacturer in Europe that needs X'. We build the set from sales calls, support tickets and lost-deal notes, and it is the single largest determinant of whether monitoring is useful.

Sample properly, because answers vary

The same prompt produces different answers on different runs, for different users, in different regions. A single run is an anecdote. We sample repeatedly and report rates with their variability rather than presenting one screenshot as a finding.

Track characterisation, not only mention

Being named as the expensive enterprise option is a different outcome from being named as the fast-to-deploy one, and neither shows up in a mention count. Characterisation is coded consistently so a shift in how you are described is visible.

Analyse the cited sources

Which pages, publications and communities the answers actually draw on. This is the most actionable output of monitoring, because it tells you where influence lives in your category — and it is frequently not your own site.

Watch competitors on the same set

Your own trend in isolation is hard to interpret when the surfaces themselves are changing. Relative movement against named competitors is a far more stable signal.

Worth knowing

Distinguish what you changed from what the model changed

Model updates shift these numbers with no warning and no relation to your work. We annotate the timeline with known provider changes and report movement against competitors as well as absolutely, because a rise that every competitor shared is not a result you achieved.

When the answer about you is wrong

There is no correction form to submit. The realistic routes, in the order we try them:

  1. Fix the source. Inaccuracies usually trace to something stated on your own site, an out-of-date third-party listing, or an old press piece that still ranks. Correcting the source is the only durable remedy.
  2. Publish the correct claim clearly, in an extractable form, on a page likely to be drawn on for that question.
  3. Correct the corroborating sources — directories, comparison sites, documentation and community answers that repeat the error.
  4. Use provider feedback mechanisms where they exist, understanding that they are slow and not guaranteed to change anything.
  5. Re-measure rather than assuming the fix propagated, because it often takes weeks and sometimes does not happen.

Where an inaccuracy is materially damaging and persistent, that is a legal question as much as a marketing one, and we say so rather than promising a marketing fix.

Process

How the engagement runs

The prompt set is built from real buyer language, because that determines whether anything downstream is useful.

Weeks 1 to 2

Prompt set and competitors

Real buyer questions gathered from sales, support and lost-deal notes; competitor set agreed.

Week 3

Baseline

Sampled runs across surfaces; presence, characterisation, cited sources and claim accuracy recorded.

Weeks 4 to 6

Monitoring build

Scheduled sampling, coding of characterisation, competitor comparison and reporting configured.

Week 7

Correction plan

Inaccuracies traced to their sources with remediation owned and sequenced.

Week 8

Handover

Reporting cadence agreed; method documented so your team can extend the prompt set.

Deliverables

What you receive

A monitoring set that tells you what is said about you, how accurately, and against whom.

01

Tracked prompt set

Built from real buyer language, extensible by your team.

02

Baseline report

Presence, characterisation, cited sources and claim accuracy across surfaces.

03

Scheduled monitoring

Sampled repeatedly with variability reported rather than single-run screenshots.

04

Competitive comparison

Share of mention and characterisation against a named competitor set.

05

Cited-source analysis

Which pages and publications the answers draw on, in priority order.

06

Correction plan

Inaccuracies traced to their sources, with remediation owned and re-measured.

Fit check

Is this the right engagement?

Worth being direct. AI Visibility and Brand Monitoring in LLMs is the wrong spend in some situations, and those are listed rather than buried.

Good fit if

  • Buyers use AI assistants to shortlist in your category.
  • You suspect AI answers misstate your capabilities or pricing.
  • Competitors appear in answers where you do not.
  • You are investing in GEO and need to know whether it works.
  • A board or executive team wants AI visibility reported alongside other channels.

Choose something else if

  • You want the content and technical work rather than monitoring. See GEO and AEO.
  • Your category is not discussed on these surfaces in any volume.
  • You want guaranteed correction of an inaccurate answer, which nobody can promise.
  • Nobody will act on the findings, in which case it is a report subscription.
Questions

Frequently asked questions

Marked up with FAQPage schema so these answers can surface directly in search results and inside AI assistant responses.

Can you tell us what ChatGPT says about our brand?

Yes, sampled across a set of prompts your buyers actually use, run repeatedly rather than once — because the same prompt produces different answers on different runs, for different users and in different regions. A single screenshot is an anecdote; rates with their variability are a measurement.

What if the answers about us are wrong?

Inaccuracies usually trace to a real source: something outdated on your own site, a stale third-party listing, or an old article that still carries weight. Correcting the source is the only durable remedy, followed by publishing the correct claim in an extractable form and fixing the corroborating sources. Provider feedback mechanisms exist, are slow, and guarantee nothing.

How is this different from social listening?

Social listening tracks what people say about you in public. This tracks what AI systems say about you when asked — which is increasingly what a buyer encounters first, and which no amount of social monitoring will reveal.

How often should it be measured?

Weekly or fortnightly for most organisations. Daily is noise given how much answers vary between runs, and monthly is too slow to catch a model update that changed how your category is described. The cadence is set to the volatility we observe in your own baseline.

Can we improve our visibility, or only watch it?

You can influence it, through extractable content, entity clarity and corroboration across third-party sources — that is GEO and AEO. Monitoring is what tells you whether the influence worked, and it is worth setting up first so the work that follows has a baseline to be judged against.

Is this the right engagement?

Tell us what you are trying to build. If a different service fits better, or if you do not need us at all, we will say so.