Over time
AI brand monitoring
A visibility check describes one moment. The answers it describes are produced by systems that change — which makes the second reading more informative than the first.
Status: recurring monitoring plans are in development and cannot be purchased. The free check and the one-off AU$99 report are available now.
A snapshot of a moving target
Traditional search reporting got away with infrequent measurement because the thing being measured held still. A page that ranked fourth on Tuesday was very likely to rank fourth on Wednesday, so a monthly report was a reasonable description of reality.
Generative answers do not hold still in the same way. The same question, asked twice, can name different businesses — not because anything about those businesses changed, but because the system sampled differently, retrieved different sources, or is simply not deterministic. That has an awkward consequence: a single reading cannot be distinguished from an outlier. You do not know whether a 61 is your position or your best day.
This is not an argument against measuring. It is an argument for measuring more than once, and for reporting a distribution rather than a figure. A visibility rate of 40% across twenty observations is a claim you can act on. “We asked once and you were not mentioned” is not.
What a baseline is actually for
The first check earns its value in a different way: it finds the things that are unambiguously broken. A noindex tag, a robots.txt blocking everything, no structured data at all, a wholly inaccurate description — none of those need a trend line to be worth fixing. Run the free check for that. The case for monitoring begins once the obvious work is done and the question becomes whether it worked.
One reading vs many
Is anything broken?
One check: Answered wellOver time: Also answeredWas this reading typical?
One check: Cannot be knownOver time: AnsweredDid my changes work?
One check: Cannot be knownOver time: AnsweredAm I gaining or losing share?
One check: Cannot be knownOver time: AnsweredDid something just change?
One check: Cannot be knownOver time: Answered
Confidence reflects the amount and quality of data available for this assessment. It is shown separately and is never folded into the AI Says Score.
Causes of movement
Five reasons your AI visibility changes without you touching anything
Understanding which cause is behind a movement is most of the value of watching it, because the four remedies are completely different.
- 01
Model updates
When a provider ships a new model version, the internal representation of your business is rebuilt from a different snapshot of the web and the answering behaviour changes with it. Businesses occasionally gain or lose mentions on the same question overnight for no reason attributable to them. This is the single largest source of unexplained movement, and it is entirely outside your control.
Remedy: none directly. Re-baseline and keep measuring.
- 02
New content — yours and everyone else's
Retrieval reflects what exists now. A competitor publishing a thorough page on a question you both want to be named for changes the selection pool for that question. So does a directory expanding its coverage of your category, or an industry publication running a round-up.
Remedy: publish against the questions you are losing, not in general.
- 03
Competitor moves
Entity work by a competitor — schema, profile consistency, corroborating listings — raises their confidence score as an entity and makes them a safer business for a system to name. Competitive share can fall while nothing about you deteriorates at all.
Remedy: match the entity hygiene, then differentiate on proof.
- 04
Changing citations
The set of sources a system reaches for shifts as pages are re-crawled, re-indexed, restructured or retired. A page that cited you last month may have been rebuilt without the mention. Your own URL changes are a frequent and self-inflicted version of this.
Remedy: keep citable URLs stable; monitor which sources replace yours.
- 05
Your own site, whether or not you noticed
A theme update that strips structured data, a staging rule that leaves a noindex in production, an SSL or redirect change that breaks crawling, a CMS migration that changes every URL. Technical regressions are common, silent, and they affect the signals that are fully within your control.
Remedy: re-run the readiness audit after every deployment.
Scope
What monitoring would track
The design principle is that a tracked metric must be capable of changing for a reason you could act on. Anything that only moves with model noise is reported as context, not as a number to chase.
Score history
The overall score and every component, plotted over time with the observation count behind each reading so a move can be judged against the confidence it carries.
Question-level presence
A stable question library re-run on a fixed cadence, so presence and absence are comparable between runs rather than compared across different prompts.
Competitive share trend
Who is named alongside you and how that mix shifts. Share is the metric where a trend is genuinely more meaningful than any single value.
Citation changes
Which sources are cited for your category, when your own domain enters or leaves that set, and which third-party page replaced you when it does.
Accuracy drift
Whether claims made about your business stay true as descriptions are regenerated. Inaccuracies tend to appear quietly and persist until somebody checks.
Readiness regressions
The twenty-four site signals re-audited, so a deployment that silently removes your structured data is caught in days rather than at the next annual review.
In development
The monitoring tiers being built
Listed so you can see the direction of travel. None of these are available to purchase, and no price has been set for any of them.
| Tier | Intended scope | Status |
|---|---|---|
| MonitorTrack your AI visibility month after month. |
| Coming soonmonthly |
| ProAdvanced monitoring for larger businesses. |
| Coming soonmonthly |
| AgencyMultiple client accounts under one login. |
| Coming soonmonthly |
Meanwhile
How to build your own trend line now
Nothing stops you assembling a useful history by hand in the meantime. Three habits do most of the work.
- 1
Fix the cadence, not the prompt
Re-run the check on the same day each month, on the same question set. Comparability comes from holding the inputs still, and a changed question library invalidates the comparison.
- 2
Record the context, not just the score
Note what you deployed, what you published and what you noticed competitors doing between runs. A score history with no change log cannot attribute anything.
- 3
Re-audit after every deployment
Site signals respond immediately and regress silently. The readiness audit is the one part of this you can verify the same afternoon.
Start from a free check and keep the results. Platform-specific behaviour is covered on the ChatGPT, Gemini and Perplexity pages.
Questions
Questions about monitoring
No. Recurring monitoring plans are in development and are not purchasable. What is available now is the free check and the one-off AU$99 visibility report. You can register interest on the pricing page, and nothing is charged for doing so.
Because a generative answer is not a fixed property of your business. Ask the same question twice and the businesses named can differ. One check establishes a baseline and tells you whether anything is obviously broken. It cannot tell you whether the result was typical, whether you are improving, or whether a change you made had any effect — all of which need at least a second reading.
Frequently enough that a real change is distinguishable from ordinary variation, and infrequently enough that the readings are not dominated by noise. For most businesses that is a monthly cadence on a stable question library, with a re-check after any substantial change to the website. Daily readings of a non-deterministic system mostly measure the system.
A sustained change rather than a single reading: disappearing from a high-intent question across consecutive runs, a new competitor appearing consistently, a factual claim about you changing, your own domain dropping out of citations, or a component score moving beyond the range of its recent variation. A one-off fluctuation is not news.
Site-level signals do — re-run the readiness audit after a deployment and the result changes at once. Observed AI answers are slower and less predictable, because they depend on re-crawling, re-indexing and in some cases on a model being updated. This lag is the main reason trend data is more informative than any individual reading.
That is the intention for the higher tiers in development, since competitive share is only meaningful as a trend. A competitor appearing once in an answer is an observation; a competitor appearing in most answers to your highest-intent questions for three consecutive months is a commercial problem with a date on it.
Set your baseline today
Run the free check, keep the result, and you have the first point on your own trend line.