Skip to main content
Reputation is how AI answers and the pages around them describe your brand when they name it, and whether they speak well of it. It reports Understood and Praised, and checks each of your claims and facts against AI answers and against the web.

​
Two parts

Understood and Praised each have an AI part and a web part. The AI part reads the answers that name you. Perception explains how. The web part reads the pages that AI cited and Google ranked for your tracked keywords and that name you. Each weekly report freezes how it read each page, so a later reread never changes a past report. Within a period a page counts once, at its newest reading, and every page weighs the same. Your own pages and your competitors’ pages are left out, and so is a page that matches your name only as an ordinary word.
  • Understood for a page is how closely it describes your category, buyer, offer and what sets you apart as you describe them. Each of the four points scores 5 when the page agrees, 3 when it partly agrees, 1 when it contradicts and 0 when it doesn’t say, out of 20. A point the reading couldn’t decide stays out.
  • Praised counts a page whose tone toward you is positive. A neutral or negative page counts as not praised. A page with no tone read stays out.
Only pages that AI cited or Google ranked count. A mention found any other way never enters either figure, so both mean the same on every plan.

​
The blend

The overview and the reputation read weigh the two parts: 0.63 × the AI part + 0.37 × the web part. These are the weights of All. The range and the change take the same weights, as if the two parts moved together.
  • Under ten pages, the web part isn’t shown and each figure is the AI part alone.
  • With no AI answer measured, a figure is the web part alone.
  • Across reports, a figure compares with its baseline only when the web part is shown in both or in neither. The report that first reads ten pages has no change.
  • Under a filter, each figure is the AI part. A page has no engine, prompt, topic or segment.

​
Facts, and what the web says

Each claim and fact carries the verdicts of the AI answers read against it, as in Perception. Reputation adds elsewhere: how many of the period’s mentions agree with it and how many contradict it. That counts every page that names you, however it was found. A page counts only at the wording it was read against, so a claim you edit starts again from the next reading. A claim’s read lists those pages, pages that contradict it first. A contradicting page shows the passage where the reading located the contradiction.

​
Read reputation through the API

GET /v1/brands/{brand_id}/reputation and MCP get_reputation return:
  • understood and praised, the blend.
  • parts, with the ai and web part of each. The web part’s answers counts pages, and its state is too_few_answers under ten.
  • facts, one row per claim and fact, with the ai verdict counts and elsewhere, its agrees and contradicts.
  • written, negative_prompts, negative_sources, best_quotes and menus, as Perception returns them.
They accept period, the shared figure filters, as_of and the list limits that Perception lists. Pass a fact’s key to GET /v1/brands/{brand_id}/reputation/claims/{claim}, or as claim to get_reputation, for its quotes, prompts, engines and contradicting pages, and documents, the pages behind elsewhere. GET /v1/brands/{brand_id}/overview and MCP get_brand_overview carry the same blend in understood and praised. GET /v1/brands/{brand_id}/scores and get_scores keep the AI part, which is also parts.ai.