AI monitoring · always on
AI is answering for your brand. We read every answer.
AI brand mention monitoring is a weekly, evidenced record of what the assistants say about you: whether you get named, how you get described, which sources they quote, and where a competitor has taken your place. Captured, dated and explained, with an alert the day it changes.
- Scored one by one, never averaged into a number that hides the bad one.
- Every assistant, separatelyScored one by one, never averaged into a number that hides the bad one.
- Every claim followed back to the page that taught it.
- Sources tracedEvery claim followed back to the page that taught it.
- Critical findings reach a human within 24 hours.
- Alerts, not dashboardsCritical findings reach a human within 24 hours.
- 30 days notice, and the data is yours to keep.
- Month to month30 days notice, and the data is yours to keep.
See it in action
One screen, everything that changed this week
- Mention share34%▲ 6pts
- Sentiment score7.8/10▲ 0.4
- New citations11▼ 2 lost
- ChatGPT78%
- Gemini64%
- Copilot52%
- Perplexity38%
- AI Overviews12%
- HighService area error repeated in ChatGPT, traced to a 2023 directory listing
- WinPricing page now cited as the source in Gemini for two prompts
- WatchFraming shifted toward “budget” in Perplexity, source under review
Sample report. The brand, the figures and the findings are illustrative — your own report carries your prompt set, your competitors and the answers captured for you.
Why it needs watching
An audit is a photograph. This is the film.
S-01 · diagnosis
A photograph, not the film
A one-off audit tells you where you stand today. It cannot tell you the week a competitor's new page pushes you out of the answer.
S-02 · diagnosis
The inputs move without telling you
Model updates, new indexed pages and fresh reviews change what gets said about you. Nothing notifies you, and there is no ranking report to check.
S-03 · diagnosis
A wrong claim repeats until its source is fixed
An old price, a closed location, a service you dropped — once a model is repeating it, it keeps repeating it to every buyer who asks.
S-04 · diagnosis
The loss leaves no footprint
A buyer who never sees you named never visits, never bounces and never appears in analytics.
What we track
Five signals, re-measured every week
All five signals are captured per assistant, so you can see that your framing is fine in Gemini and wrong in Copilot rather than averaging the two into a number that hides both.
- 01
Mention frequency
How often you are named across a fixed prompt set, tracked as a line over time rather than a single number — so a slide shows up as a trend, not a surprise.
- 02
Framing and sentiment
The adjectives the models attach to you — premium or cheap, specialist or generalist, reliable or unproven — and which source taught them that.
- 03
Factual accuracy
Every claim checked against the truth: locations, pricing, services, ownership, credentials. Errors are traced back to the page teaching them.
- 04
Cited sources
Which URLs the answer is built from — yours, a competitor's, a directory, a forum thread — and when that set changes underneath you.
- 05
Competitive share
Your share of named recommendations against the same competitor set each week, so movement is comparable and not re-baselined to flatter anyone.
Alerting
We contact you for four things, and nothing else
Monitoring that emails you daily gets muted in a fortnight. These thresholds are agreed with you at setup and tuned after the first month.
- Critical
A false or damaging claim appears
Wrong pricing, a service you don't offer, a legal or safety implication, or a confusion with another company. We send the captured answer, the cited source and a recommended correction path.
You hearWithin 24 hoursPhone or Slack, not email - High
You disappear from a prompt you held
A prompt that named you last week no longer does. We identify what replaced you and whether it is a new competitor page, a new cited source or a model change.
You hearSame weekWith the diagnosis attached - Watch
Your framing shifts
The language used about you moves in a direction you would not choose. Usually traceable to one review, one article or one outdated page — and usually fixable at the source.
You hearWeekly digestEscalated if it persists - Win
You gain ground worth repeating
A page of yours becomes the cited source, or you enter a prompt set you were absent from. Worth knowing, because it tells you which work is paying and should be repeated.
You hearWeekly digestWith the cause identified
Platforms monitored
Eight assistants. They do not agree about you.
Each one is prompted, captured and scored separately every week, because a brand can be the default answer in one and invisible in another.
- ChatGPTBrowsing & non-browsing
The highest-volume assistant, and the one most buyers mean when they say “I asked AI”.
- ClaudeWith web search
Heavily used inside businesses, which makes it disproportionately important for B2B shortlists.
- GeminiApp & in-Search
Where Google's own properties and your structured data carry unusual weight in who gets named.
- PerplexityCitations logged
The most source-transparent surface, and the fastest read on which pages decide your category.
- Google AI OverviewsPanel & AI Mode
Seen by people who never opened an AI app at all, the largest passive audience of the eight.
- Microsoft CopilotCitations logged
Bing-fed and built into Windows and Office, which matters more in enterprise research than its share suggests.
- GrokSocial-weighted
Leans on live social conversation, so it surfaces reputation problems before the other seven do.
- Meta AIConsumer & local
Inside WhatsApp and Instagram, where consumer and local-service questions actually get asked.
DeepSeek, Mistral's Le Chat and any regional assistant your buyers use can be added to the same prompt set on request. All platform names and logos are the property of their respective owners and appear here only to identify the surfaces monitored.
What lands, and when
No dashboard nobody logs into. Everything arrives where your team already works.
- As it happensCritical alerts
Slack, Teams or phone. Captured answer, cited source, recommended action — no dashboard login required.
- WeeklyThe digest
One page: what changed, what caused it, what we are doing about it. Readable in two minutes by someone who is not a marketer.
- MonthlyTrend report
Mention share, framing and competitive position plotted over time — same prompt set, same competitors, every month.
- QuarterlyReview & re-scope
A live session: what moved, what the prompt set should become next quarter, and which corrections are still open on your side.
When something is wrong
You cannot edit the model. You can change what it reads.
Every correction works the same way: find the source the answer was built on, then change it or outweigh it.
- 01
Find the source
Identify the page teaching the claim — a directory entry, an old review, a stale press release, or one of your own forgotten pages.
- 02
Fix what you own
Update the page, its schema and your business profiles, so the accurate version is the freshest and best-structured one available.
- 03
Outweigh what you don't
Request corrections from third-party sites; where that fails, publish and earn coverage strong enough to outweigh the bad source.
- 04
Re-run until it moves
The prompt goes on a watchlist and is re-run weekly until the answer changes — and you are told either way, including when it does not.
Corrections are not instant. Models re-read the web on their own schedule, so a fixed source typically takes weeks to show up in answers. We report the fix date and the date the answer actually changed, separately — nobody gets to claim credit for a change that never happened.
Scoping
Three levels, priced on your prompt set
Cost follows prompt volume, assistant count and how many markets and languages you sell in. Fixed monthly fee, quoted before anything starts.
- Essential
Watch the brand
For a single brand in one market that mainly needs to know when something is wrong.
- 50 prompts, re-run weekly
- 4 assistants
- Critical alerts and weekly digest
- Monthly trend report
- Corrections advised, run by your team
- Most chosenMost chosen
Watch the category
For brands that also need to see competitors moving, and want the corrections handled.
- 200 prompts, re-run weekly
- All 8 assistants
- Up to 5 competitors tracked
- Source tracing and correction work included
- Quarterly review session
- Multi-market
Watch everything
For multi-location, multi-language or multi-brand groups where each unit needs its own view.
- Prompt set per market or brand
- All assistants, plus regional ones
- Language-specific capture and scoring
- Per-unit reporting plus a group roll-up
- Named lead and monthly working session
Straight answers
Before you ask us
Answer
Can't we just check this ourselves?
You can, and you should occasionally. What is hard to do yourself is consistency: the same prompts, the same accounts, the same wording, captured on the same day each week across eight assistants, without your own logged-in history skewing the answers. Ad-hoc checks tell you what one person saw once; monitoring tells you what changed.
Ask us something harderAnswer
How is this different from social listening?
Social listening watches what people publish about you. This watches what machines say about you when nobody is publishing at all — a synthesised answer, delivered privately to one buyer, that leaves no post, no mention and no trace in any listening tool. Different surface, different mechanics, different fix.
Ask us something harderAnswer
Can you force a model to correct something?
No, and neither can anyone else. There is no edit button and no support queue for what a model asserts. What works is changing the sources it reads and adding stronger, fresher, better-structured ones — then waiting for the re-read. We report both dates so you can see how long it genuinely took.
Ask us something harderAnswer
Do you use an off-the-shelf tool for this?
We use tooling for capture and storage, and people for the part that matters — reading the answers, judging whether a framing shift is material, and tracing a claim to its source. A tool can tell you a number moved. It cannot tell you that one review from 2023 is the reason.
Ask us something harderAnswer
What if the models rarely mention us at all?
Then monitoring is the wrong first purchase. Start with the competitor analysis or the visibility audit to establish why, do the work to get named, and add monitoring once there is something to protect. We will tell you that on the call rather than sell you the recurring fee.
Ask us something harderAnswer
Is there a contract?
Month to month with 30 days' notice, and the prompt library plus every captured answer is yours to keep if you leave. The first month includes setup — building the prompt set, agreeing the competitor list and tuning alert thresholds — so month one is the heaviest and cheapest to walk away from.
Ask us something harderMonitoring tells you. These change it.
All AI & search services- BenchmarkCompetitor AI Visibility AnalysisThe one-off benchmark most clients run before starting monitoring.
- DiagnosticAI Visibility AuditYour own visibility in isolation, with a fix list attached.
- ProgrammeAI Search VisibilityThe programme that gets you named in the first place.
- SourcesReputation ManagementFix the reviews and coverage the models are reading about you.
- StructureGenerative Engine OptimizationRestructure pages so the models can quote you accurately.
- AuthorityDigital PR & Link BuildingEarn the third-party sources that outweigh a bad one.
Find out what they are saying this week.
Send your brand name, your category and one competitor. We will run a free sample across four assistants and send back the captured answers — including anything factually wrong — before you decide whether to monitor at all.
Monitoring runs month to month with 30 days' notice. No annual lock-in, because a service you would leave in month three should not be able to hold you for twelve.
