Scorecard

What the scorecard is

The scorecard (Performance → Scorecard) tracks a set of KPIs across five families: crawl and retrieval, brand queries, non-brand queries, the paid signal from ChatGPT ads, and business outcome. Each KPI has a quarterly target. The targets are set once, when a baseline month is locked, and do not move afterwards: every quarter is judged against the number that was frozen.

Every target is a projection, and the page says so beside each one. A target is a planning figure derived from the baseline and a small set of factors, not a forecast of what will happen.

Two crawl rows read from your server logs once they're connected. Priority-page crawl coverage is the share of the pages you list under Figures you enter that an AI search crawler reached in the month. AI-bot error and block rate is the share of AI-bot requests that got a 403, a 404 or any other error; a redirect isn't a failure. Challenge pages and firewall blocks appear only if your CDN's logs are connected.

The tabs

• Quarterly scorecard: one row per KPI with its baseline and its four quarters. Each quarter shows its target, the actual, and a state: on track, close, behind, within noise (the difference is smaller than the measurement can tell apart), so far (the quarter is under way: the average of its finished months, or the month in progress's reading when none has finished, updated nightly and shown beside the target with no colour and no judgement), awaiting a reading (the quarter has not started, with the day of its first reading, such as "First reading from 1 Feb 2027"), or not set (there is no target). A quarter is judged, and takes its colour, once it has ended. The counts behind each rate (how many answers to how many questions) are printed with it. A strip at the top sums each family, counting the measures awaiting a reading apart from those with no target.
• Monthly progress: the same KPIs month by month, with each month's target, reading and state. The month in progress is marked open.
• Competitors: the three rivals named most often alongside you in the baseline month, with your share and theirs and how far you are behind the leader.
• Target logic: how each target was set: the KPI's type and evidence grade, how fast it is expected to move, its limit, and the factors used. Read-only.
• Business impact: the value calculator, from four figures you enter yourself: enquiries per month, the share influenced by AI, your close rate and your average deal or order value. Its results are worked out from your figures, in your own currency; we do not project revenue. Only your organisation's owners and admins see this tab's figures.
• Setup: the locked baseline: its month, the position tier and scenario chosen, the market, and the number of questions frozen with it.

Open negative or inaccurate claims

This row reads by itself when you haven't entered a figure for a month. It counts the facts stated wrongly in two or more answers on brand questions that month, each fact once, from the accuracy checks we already run, so it costs nothing. A figure you enter for a month wins, because your own log can hold more than we can read: it counts wrong facts only, not negative statements. The three standard questions about who you are aren't counted, as they aren't in any figure here. A month with no checks shows no reading, never zero: checks need your brand facts.

How this builds value over time

The Business impact tab, and the page before a baseline is locked, also carry the method's own words on how the work pays off: three horizons (the first two quarters on the retrieval clock, quarters three and four mixed, beyond the first year on the memory clock), why earned visibility compounds, and the internal checks. Each keeps its evidence grade, so a projection reads as one. They hold no figures, so everyone with access to the project can read them.

Where we differ from the method. The method collects through the logged-out interface, and notes that API and interface sources have overlapped as little as 8% on Perplexity. We ask the chat engines through their APIs, with live web search on, because a call through the API can be repeated, priced and stamped with how it was answered, and the logged-out interface can't at this scale. So the sources behind a rate here can differ from the ones a person sees in the engine's app. Google AI Overview, Google AI Mode and Bing Copilot are read from the search engine's own page.

Which questions a KPI is read on

Neutral answers with no wrong fact. Beside the two favourable shares (B3 and C6), a line gives the share of the same judged answers that were neutral and had no wrong fact. It reads what the tone check and the fact check already recorded, and it is not counted in the favourable share. A neutral answer with no check result never counts as accurate, whether nobody checked it or the check found nothing it could verify, so the line can understate, and it says how many there were. With no fact check in the month there's no line, rather than a 0.

The switch at the top chooses between the Baseline cohort, the questions frozen when the baseline was locked, and the Full set, the questions you track today. On the full set, each row is judged on the questions it tracks today, while its target was set from the questions frozen at the lock. A few KPIs read from your server logs or Google Analytics, which know nothing of questions, so they always read the full set, and the page says so beside them. The two counts of queries tracked (B1 brand, C1 non-brand) always read the full set too: their targets grow as you add questions, and the baseline cohort never does. Facts stated correctly (B2) always reads the full set as well: the fact check is counted across every brand answer. It is the same figure as the Accuracy column on Platform comparison, and it needs brand facts to check answers against.

A lock freezes only what it could read that day. A measure that begins later, a source connected later, a figure entered later, or a rate that was under the floor at the lock gets its baseline and targets automatically, from the first complete month it can be read. They fit the lock's own quarters, nothing already frozen moves, and the row says so: "Added 4 Nov: measured from Oct 2026".

Scenarios

Targets are frozen for three scenarios: conservative, base and stretch. The page opens on the scenario chosen at the lock; the scenario switch shows how the same readings stand against the other two.

In reports

Standard, comprehensive and AEO Program reports carry a Quarterly Scorecard section: the same view as this page, on the baseline cohort and the scenario chosen at the lock. The PDF shows each KPI's quarters and the competitors; the Excel file adds monthly progress and the target logic. Q1 to Q4 count from the baseline month, whatever dates the report covers, and the section is not filtered by a report's scope. Before a baseline is locked, it says so in one line. The value calculator is never in a report: it uses your organisation's confidential sales figures, and a report can be read by anyone on the project or through a share link.

Locking a baseline

Your organisation's owners and admins lock a baseline on the Setup tab; anyone else sees who can. The form asks for:

• Baseline month: a whole month of captures in the last 12 months. Each month says what it holds on the questions the lock would freeze: "Oct 2026 · 396 answers · 1 run", or "partial runs only" when no capture run completed. A run counts in the month it started, on your project's time zone, the same calendar its answers are dated in. A month with no answers on those questions is shown but cannot be chosen, and the latest month that holds answers is chosen for you. When no complete month holds answers yet, the form says so and names the day the month in progress can be locked.
• Position tier: suggested from your Share of Model in that month, or one of the five tiers, each with the factor it applies. The suggestion is read for the month you choose and names the tier it would pick ("Suggested from your Share of Model: Competitive (23.4%)"). When Share of Model has too few answers that month to set a baseline, or cannot be read just now, the suggestion cannot be chosen and you pick the tier yourself before locking.
• Scenario: conservative, base or stretch. None is chosen for you: the scenario is a decision.
• Market: the project's home market, kept with the baseline. It is chosen on Locations, not here. With more than one country and none chosen, you can choose it first or lock without one. A later choice leaves this lock as it is.
• Overrides (optional): a different ceiling or floor for a KPI, only when the category is unusually narrow or crowded, each with its reason.

Before anything is saved, the page asks you to confirm, because the targets freeze: every quarter is then judged against them. If the lock is refused, the page shows the reason exactly as given (a month outside the window, a missing reason, another lock that finished first).

Re-baselining replaces the frozen targets with new ones and needs a reason; the old baseline is kept and linked to the new one.