# Audit Audit answers two questions about a site, on two lenses over the same evidence. - **Health** — what is wrong. Every open issue, grouped into families, each leading with what it is costing. - **Opportunities** — what it is worth. The detector families with their own numbers and the visits on the table. Switch between them on the rail under the page title. The lens is in the address, so a link to `?lens=opportunities` opens where you left it. Nothing on this page starts a crawl or spends anything. Everything is read from evidence already stored. ## Running an audit SERPclimber audits every project automatically about once a week. The status line beside the page title always says where the last run got to: - **Crawled Sep 2 · 1,240 pages** when nothing is running. - **Crawling · 412 of 1,240 pages** or **Analyzing · 1,240 pages** while a run is live, counting real pages as they are fetched. - **Last audit failed**, with the reason, when something went wrong. Choose **Run audit** to start one now, or **Cancel** to stop one that is running. A cancelled run leaves the previous findings untouched. An audit crawls the site and inspects crawl access, status and fetch failures, canonicals, redirects and noindex directives; titles, descriptions, heading order, thin content, image alt coverage and structured data; internal links, orphan risk and pages buried four or more clicks deep; AI crawler access; and the directory and profile catalog. Repeated observations update the same finding rather than creating noise, and repeated page-level findings are grouped into one issue with every affected page attached. ## Plain-language explanations Every finding carries three lines written for someone who has never read an SEO article: - **What it is costing** — the summary line on the row, using the finding's own numbers: "Google shows /running-shoes 2,400 times a quarter and about 1 in 100 people click — fewer than pages in the same positions normally get." - **What fixing it is worth** — shown when the row is expanded. - **What to do** — the next concrete move, also on the expanded row. The lines are written by the analysis model from the finding's own evidence, and fall back to a per-family template when no model is configured or a response cannot be used. The numbers always come from the evidence, never from the model. When a finding becomes a task, its explanation travels with it. ## Health lens Rows are grouped into **Critical & high**, **Medium**, and **Low & info**, one card per band. Each row shows its title, what it is costing, the search demand behind it as a number and a bar, how many pages it affects, when it was first seen, and whether it has grown or shrunk since the last run. Chips carry the severity, the finding type in plain words, the department, and the state. Select a row to expand it: what fixing it is worth, what to do, and the full list of affected pages. **The verbs.** On any open row: - **Create task** — the whole family, every affected page, becomes one task. Later pages joining the family join the existing task. - **Snooze 7 days / 30 days** — the row disappears from **Open** and comes back on its own when the date passes. Nothing is lost; snoozing is for "not now", not "not real". - **Mark resolved** — you fixed it outside SERPclimber. The next audit reopens it if it is still there. - **Ignore** — it is not a problem for this site. It stays ignored until the underlying evidence materially changes. Tick the checkboxes to select several rows and use **Create tasks** or **Ignore** on all of them at once; each family still becomes its own task. **Filters.** Department chips on the left narrow the list to Site health, Search visibility, Content, Links & authority, AI visibility, Local SEO, or Analytics & tracking. State chips on the right switch between **Open**, **In task**, **Snoozed**, **Resolved**, **Ignored**, and **Any state**. Open is the default: findings already in a task live on the Tasks page. ## Opportunities lens One card per family, each ranked by what it is worth, with the numbers that family actually reports: - **Content decay** — visits now against the earlier baseline and the drop, with a 16-month strip per row so a slide is visible at a glance. - **Keyword cannibalization** — how often the search is shown, how many of your pages compete for it, and the best rank among them. - **Striking distance** — the visits to gain, how often you are shown, and where you rank. - **Click-through gap** — how often you are shown, what share clicks, what is normal at that rank, and the visits missed. - **Authority gap** — sites linking to you against the competitor median, your rank, and the number of link targets already linking to competitors. - **Lost links** — the site, its rating, and when the link went. Every row's summary line is its plain-language cost, every column sorts, **Expand** shows the whole family rather than the top five, **Create task** turns a row into work, and **Explain with AI** hands the card's rows and numbers to the assistant. ## From finding to task Choosing **Create task** moves the family into one task with every affected page attached and its explanation copied across. Resolving the task does not erase the finding: Audit keeps the evidence trail and reopens the finding if verification fails, the problem returns, or a link disappears. ## In chat Ask "what's wrong with my site" or "what should I fix first" and the assistant reads this same model, and answers in the same plain-language lines you see here.