Stop reading dashboards. Start working a queue.

By a single published formula: score = promptsAffected × revenueWeight(cluster) × gapSize × confidence ÷ effortCost. Prompts affected and gap size come from your tracked prompt runs, confidence is the detector’s own score after any staleness penalty, and effort cost is derived from the estimated hours. revenueWeight is 1.0 until a CRM is connected, at which point clusters that resolve to open pipeline start outranking clusters that do not.

No card is emitted. A recommendation has to name at least one source and the date that source was read, or it does not reach the queue — there is no fallback that fills in a plausible-looking justification. Where evidence exists but is old, the card is still shown and its confidence is reduced, with the read date on the card so you can see why.

No. Every module the queue routes to requires a human to approve the specific artefact before it ships — the page change, the outreach email, the draft, the campaign. That includes paid: campaigns are drafted with targeting and caps prefilled and then wait for review. Nothing in the queue acts on its own.

A baseline is frozen when the card is created: the prompt cluster it covers, the mention rate across that cluster, the run count behind it, and the scorer version. When the work ships, checkpoints are taken at 7, 14, 30 and 60 days against that same prompt population. A window with too few runs to resolve the predicted effect is reported as thin rather than averaged into the result.

The queue re-orders. Targetlytics syncs HubSpot, Pipedrive, Attio and Zoho and identifies the companies arriving from AI answers, so a prompt cluster can be tied to named accounts and their open deals. A cluster carrying pipeline outranks a cluster with more prompts and no commercial weight behind it. Nothing else about the cards changes — the same findings, in a different order.

An audit grades pages against a checklist and hands you the failures. This queue starts from what AI answers actually did — which sources were cited for your prompts, where your brand was absent, what moved — and it names the module that carries each item out. An audit finding is true about your site; a card here is true about your answers.

They are Targetlytics’ own queue, read on 6 September 2026 and put through the same filters every customer queue gets: the fan-out collapsed, ineligible targets dropped, priority-tier impact figures removed. Nine pitch targets that were live in this queue that morning — including usa.gov, Yelp and doi.org — are not on this page because none of them passes the eligibility gate.

Yes, and a dismissal is recorded rather than discarded. A cluster you chose to leave alone is a natural control group: it lets the effect of the work you did do be measured against comparable clusters where nothing changed.