Skip to content
Back to search

now-visibility-progress-check

Monthly progress check for a site that already has a NoW search visibility audit. Re-scores the site on the same scoring model (the v7.0 AI Search Readiness Index), compares it with the original audit like for like on the checks assessed both times, sorts every earlier finding into fixed, started, unchanged or new, tracks the 30/60/90 plan, re-runs the same frozen AI buyer prompts per engine against a competitor control set, and delivers a designed, chart-led progress PDF in the NoW visual system (opening with a plain-words SEO / AEO / GEO explainer). Use when someone asks for a "monthly update", "progress check", "what's changed since the last audit", "re-run the audit and compare", "how are we tracking on SEO / AI search", or "show the progress since [date]". For a first audit with no baseline, use now-search-visibility-audit instead.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/now-of-work--now-visibility-progress-check/badge.svg)](/listings/now-of-work--now-visibility-progress-check)

HTML

<a href="/listings/now-of-work--now-visibility-progress-check"><img src="/api/listings/now-of-work--now-visibility-progress-check/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

**Claude desktop app or claude.ai**

1. Download `now-visibility-progress-check.zip` from the [latest release](https://github.com/NoW-of-Work/now-visibility-progress-check/releases/latest). 2. In Claude, open **Settings → Capabilities → Skills** and click **+**. 3. Upload the ZIP. Install the audit skill the same way; this skill uses it for the re-score.

**Claude Code**

git clone https://github.com/NoW-of-Work/now-visibility-progress-check.git ~/.claude/skills/now-visibility-progress-check

Requirements

- Web fetch and web search, and the NoW Search Visibility Audit skill (v7.0 or later). - Node.js 18 or later (no packages to install) and any headless Chromium for the PDF: Chrome, Edge, Chromium, Brave, or Python Playwright. Without one you get a print-ready HTML file. - Optional: OpenSEO, Google Analytics or Search Console connectors for sharper numbers.

---

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 246 lines
  1. ---
  2. name: now-visibility-progress-check
  3. description: >
  4. Monthly progress check for a site that already has a NoW search visibility audit. Re-scores the site on the
  5. same scoring model (the v7.0 AI Search Readiness Index), compares it with the original audit like for like on
  6. the checks assessed both times, sorts every earlier finding into fixed, started, unchanged or new, tracks the 30/60/90 plan, re-runs
  7. the same frozen AI buyer prompts per engine against a competitor control set, and delivers a designed,
  8. chart-led progress PDF in the NoW visual system (opening with a plain-words SEO / AEO / GEO explainer). Use
  9. when someone asks for a "monthly update", "progress check", "what's changed since the last audit", "re-run the
  10. audit and compare", "how are we tracking on SEO / AI search", or "show the progress since [date]". For a
  11. first audit with no baseline, use now-search-visibility-audit instead.
  12. license: CC BY 4.0. See LICENSE.
  13. compatibility: >
  14. Needs web fetch and web search, plus the now-search-visibility-audit skill for the re-score. Node.js 18+ and
  15. any headless Chromium (Chrome, Edge, Chromium or Python Playwright) for the PDF; without one it delivers a
  16. print-ready HTML file. Optional connectors sharpen the numbers: OpenSEO, Google Analytics, Search Console.
  17. metadata:
  18. author: The NoW of Work
  19. version: 1.2.1
  20. ---
  21. # NoW Visibility Progress Check
  22. You compare a site's search visibility now against its last NoW audit and show, in one designed report, what
  23. moved, what didn't, and what to do next. The reader is a business owner or their team, not an SEO specialist:
  24. every page leads with the finding as a sentence, and the report opens by explaining SEO, AEO and GEO in plain
  25. words, because most readers don't know the difference.
  26. The check is only useful if the two points in time are measured the same way. Keep the scoring model, the
  27. depth and the buyer prompts identical to the baseline wherever you can, and say plainly where they differ.
  28. Canadian spelling throughout. No em dashes where a comma, colon or full stop works.
  29. ### Files in this skill
  30. | File | Read it when |
  31. |---|---|
  32. | `references/progress-method.md` | Steps 4–6. How to classify findings, the data file field by field, comparability rules. |
  33. | `assets/report/render-progress.mjs` | Step 6. Builds the report from the progress data file. Run it; don't edit it per report. |
  34. | `assets/report/example-progress-v7.json` | Step 6. The worked example on scoring model 7.0. Copy this shape. |
  35. | `assets/report/example-progress.json` | Local only, not in the public repo (it holds real client data). Only to re-render an old v6 progress report; never copy it for a new check. |
  36. The design files in `assets/report/` (`report.css`, `charts.mjs`, `components.mjs`, `scoring.mjs`,
  37. `to-pdf.mjs`) are a copy of the audit skill's. If you change one, change both.
  38. ---
  39. ## Step 1: Find the baseline (locate it; don't open its scores yet)
  40. Look where audits are saved (a folder the user names, `$AUDIT_OUT`, `/mnt/user-data/outputs`, or
  41. `./visibility-audits/`) for:
  42. 1. the **original audit's** computed scores, `now-visibility-audit-<domain>-<date>.scores.json` (written by the
  43. audit renderer: model version, index, both ranges, pillars, checks met per discipline, every gate and check
  44. result, per-engine visibility and the competitor control set), and its data file
  45. `now-visibility-audit-<domain>-<date>.json` (findings, frozen prompt panel, plan). Its scores and visibility
  46. stay the baseline (`scores.baseline`, `visibility.baseline`) for every monthly check and every quarterly
  47. re-audit. Don't roll the baseline forward each month: comparing only with last month
  48. hides slow gains, and testing every month against a moving baseline produces false "real change" calls;
  49. 2. the last progress data file, `now-visibility-progress-<domain>-<date>.json`: its open findings and plan
  50. tracker are where this month's classification starts (Step 4). Don't use its `scores.current` as the baseline;
  51. 3. the history file `visibility-history-<domain>.json` (every past score, for the trend line).
  52. **Find these files, but don't open the scores yet.** The baseline `.scores.json`, the `score` block and findings
  53. in the baseline data file, and the last progress file all show the old check results, and the re-score in Step 2
  54. must be done before you see them. From the baseline data file, read only the frozen inputs the re-score needs:
  55. the depth, the date, the key pages and the prompt panel's wording, sources, engines, location, language and
  56. competitor names. One way to print only those:
  57. ```bash
  58. node -e 'const A=require(process.argv[1]);console.log(JSON.stringify({meta:A.meta,keyPages:A.keyPages,protocol:A.visibility?.protocol,panel:(A.visibility?.panel||[]).map(p=>({prompt:p.prompt,intent:p.intent,branded:p.branded,source:p.source,engines:Object.keys(p.results||{}),competitors:[...new Set(Object.values(p.results||{}).flatMap(c=>Object.keys(c.competitors||{})))]}))},null,2))' "<path>/now-visibility-audit-<domain>-<date>.json"
  59. ```
  60. If only a PDF exists (older audits were Word-based or v6), read it with `pdftotext -layout` and rebuild the
  61. baseline from its findings: every finding and its status, the buyer prompts and results, and the 30/60/90 plan.
  62. An old PDF carries a v6 AI Visibility Score, which can't be compared; rescore the findings on v7 (below). Tell
  63. the user the baseline came from the PDF. A PDF shows the old results beside the key pages and prompts, so if you
  64. have to read it before the re-score to recover them, the re-score isn't blind: record `meta.blindRescore: false`.
  65. If there's no baseline at all, stop and offer the full audit instead: a progress check needs something to
  66. compare against.
  67. Confirm in one line: "Comparing against the original [date] [Quick/Full] audit. Same depth this time?"
  68. Default to the same depth without waiting if the user has already said "monthly update". (Give the baseline
  69. index and range after Step 2, once you have opened the scores.)
  70. **Model version.** Audits from v7.0 on use the AI Search Readiness Index. Both sides of the comparison must
  71. carry the same scoring model version **in their score data** (`modelVersion` in each `.scores.json`, copied into
  72. `scores.baseline` and `scores.current`); the renderer refuses a mismatch or a missing version, whatever `meta`
  73. says. A v6 baseline (the old AI Visibility Score) must be rescored on v7 from its findings first: mark checks
  74. you can't reconstruct `unverifiable`, write a v7 `.scores.json` for it, and say so on the method page. "Same
  75. model" means the scoring model; changes to the AI engines' own models can't be frozen and are logged as platform
  76. events.
  77. ## Step 2: Re-score the site (blind), then open the baseline
  78. Run the audit with the **now-search-visibility-audit** skill at the same depth as the baseline, following its
  79. Steps 2–4 (crawl, analyze, score) and its Step 6 data file, using the frozen key pages and prompt panel from
  80. Step 1. **Score blind:** fill in every check and gate result for this month before you open the baseline
  81. `.scores.json` (or the baseline data file's score block, or the last progress file). The cleanest way is a fresh
  82. sub-agent given only the site, the key pages and the panel. Record `meta.blindRescore: true` in the progress file,
  83. or `false` if you saw the baseline results first. Save the new audit data file and its `.scores.json`; they become
  84. this month's `scores.current` and `visibility.current`, while the baseline stays the original audit.
  85. **Then open the baseline.** Only now read the baseline `.scores.json`, the baseline findings and the last
  86. progress file (Step 1, items 1 to 3). The findings may be read freely from here on, for the "what we fixed" story
  87. and Step 4. Compare the two sets of results: the renderer lists every check that flipped, and for each partial ↔
  88. met or not-met flip, record the evidence for the new call in the method caveats, so a judgment change isn't read
  89. as progress. You don't need to deliver the full audit report unless the user asks for it; the progress report is
  90. the deliverable.
  91. If the user already ran the audit and gives you the file or PDF, use it instead of re-crawling.
  92. ## Step 3: Monthly measurements
  93. Do these every month, in the same way each time, and only report what you actually measured:
  94. - **Buyer prompts.** Re-run the frozen prompt panel saved in the baseline (`visibility.panel`): same wording,
  95. sources, engines, location and language, and at least the baseline's runs (8 or more per prompt per engine),
  96. spread over at least 8 separate dates (distinct dates are counted; in practice one run a day for 8 days).
  97. Record `runs`, `mentions`, `searched`, `days[]` and `competitors` per prompt per engine, in the audit data
  98. file's panel shape, listing every competitor in every cell and writing `0` explicitly when none of the runs
  99. named it (a missing name is counted as 0 without warning), and render the audit so its `.scores.json` carries the
  100. per-engine rates and the competitor control set. Use the same competitors as the baseline, at least three of
  101. them: with fewer, the report labels the control "indicative". If direct AI queries are blocked, use the
  102. search-proxy tier and say so; a proxy result is never compared with a measured one, and the renderer refuses
  103. the pair.
  104. - **What the panel can see.** At 20 prompts over 8 days the re-check only detects large moves (at least about
  105. 16 points; each report prints its own figure). Precision comes from more prompts run on more separate days, not
  106. from more runs of the same prompt on the same days. For month-to-month tracking use 40 or more prompts, or treat
  107. monthly results as descriptive and test quarterly. Say so to the client before the first monthly check. Monthly
  108. checks are descriptive (`meta.formalTest: false`, the default); set `meta.formalTest: true` only at the quarterly
  109. re-audit, which is when "real change" verdicts are given. Even quarterly testing carries some risk: each verdict
  110. has a 5% false-alarm rate, so over a year of quarterly tests there is about a 1-in-6 chance of one false "real
  111. change" when nothing changed. Tell the client, and treat a single real-change verdict as provisional until the
  112. next quarter agrees.
  113. - **Index checks.** Search the brand on Brave and on DuckDuckGo (Bing) and note whether the new pages appear
  114. and the stale ones are gone. These are the threshold fixes most audits flag.
  115. - **Connected data, if available.** OpenSEO (AI visibility trend, rank tracker), Google Analytics (the AI
  116. referrals channel), Search Console (clicks and impressions for the pages that changed). Use them to add facts,
  117. never to fill a gap with a guess. If none is connected, say what each would add.
  118. ## Step 4: Classify every finding
  119. Read `references/progress-method.md`. Go through every finding from the baseline and sort it into **fixed**,
  120. **started**, or **unchanged**, each with the evidence you saw this month. Then list what is **new**. Mark the
  121. status of every row in the baseline's 30/60/90 plan (done, in progress, not started, overdue).
  122. Three rules matter most:
  123. - **Only call something fixed if you saw the fix.** Quote the URL, the page section, the count.
  124. - **Separate real change from measurement change.** If the scan was deeper, a page wasn't checked last time, or
  125. a login turned an unverifiable check into an assessed one, say so in the method caveats rather than counting it
  126. as progress or decline.
  127. - **Call a change real only when it beats the noise** (full rules in `references/progress-method.md`):
  128. - **Index**: the renderer recomputes both totals like for like, over checks assessed in both audits, and
  129. prints the raw move beside it as "including measurement change". Report the checks whose result changed. A
  130. move driven only by newly verifiable checks is labelled "measurement change".
  131. - **Mention rate, per engine**: at the quarterly re-audit (`meta.formalTest: true`), a change is real only if
  132. it survives a two-proportion test on the cluster-robust standard errors with a Holm correction across
  133. engines, after subtracting the median competitor change on the same engine. The test treats the two waves as
  134. independent samples although they share the frozen prompts, which makes it conservative: it can miss a small
  135. real change, but it doesn't invent one. Otherwise "within measurement noise"; if the competitors moved the
  136. same way, "platform drift". In monthly checks the renderer shows the moves as "Descriptive" and gives no
  137. verdict.
  138. - **Outcome counts** (GA4 AI sessions, Bing citations, Search Console AI impressions): real only if the change
  139. exceeds 2 × √(2 × baseline count) (about 17 for 38 sessions) and holds for two consecutive months, or exceeds
  140. 2 standard deviations of at least three pre-period months when you have them.
  141. ## Step 5: Write the story
  142. Each page leads with one headline that states the finding (`*text*` marks the teal half). Write them last,
  143. once the numbers are in. The cover headline says what changed and what didn't in two short sentences. Be
  144. straight: a flat or falling month gets a plain headline and the reason, not spin.
  145. ## Step 6: Build the report
  146. Write the progress data file (`references/progress-method.md` has every field; copy the shape of
  147. `example-progress-v7.json`) and run:
  148. ```bash
  149. node "<skill-dir>/assets/report/render-progress.mjs" "<out>/now-visibility-progress-<domain>-<YYYY-MM-DD>.json" "<out>"
  150. ```
  151. It refuses the file, with a plain message, when: either score block lacks `modelVersion` or the two differ;
  152. `meta.modelVersion` disagrees with the score data; a v7 score block lacks `pillars`, `disciplines`, `checks` or
  153. `gates` (a typed or relabelled block); a v7 file carries the old `citation` block; the two visibility tiers
  154. differ; or a measured engine has no standard error (`se`), which means the audit was rendered with an old
  155. `scoring.mjs`. It accepts a range-only baseline (no point estimate, `total: null`), and it decides whether either
  156. side is capped from its gates (any `fail`), not from comparing `total` with `uncapped`.
  157. It produces, in order: cover (before → now index, like for like on checks assessed both times, on the band
  158. track; SEO / AEO / GEO checks met then and now), **Start here: SEO, AEO and GEO explained**, scorecard (trend line
  159. once there are three or more months, pillar dumbbells for pillars with checks assessed both times, each side's
  160. unverified range, which checks changed, the waterfall of where the points came from, with a "Gate cap" step when a
  161. failed gate capped either side), what we fixed, where every finding stands, still open with the AI visibility
  162. check (per engine: before and now with ranges, competitor drift, net change, the baseline's detectable change,
  163. and the verdict or "Descriptive"; engines run in only one audit are listed as not compared; a proxy pair gets one
  164. "named in X of N checks then, Y of M now" line), plan tracker, next steps, then the carried-forward proof,
  165. off-site and 90-day pages if you include them, and the method.
  166. Render check: rasterize the cover, scorecard and one list page (`pdftoppm -r 60 -png`), look for text running
  167. off a chart, an awkward headline break or an empty section, fix the data and render again. The cover must fill
  168. the whole page; if it doesn't, a chart is wider than its column.
  169. ## Step 7: Save the trail
  170. 1. Append this month to `visibility-history-<domain>.json`:
  171. `{"label": "19 Dec", "date": "2026-12-19", "modelVersion": "7.0", "total": 63, "range": [60, 66], "visibility": {"tier": "measured", "engines": [{"engine": "ChatGPT", "rate": 15}, {"engine": "Perplexity", "rate": 12}]}}`.
  172. `total` is that audit's raw index from its `.scores.json` (null when it had no point estimate); the trend line
  173. plots it. The cover's like-for-like figure changes with every pair of audits, so it isn't stored here.
  174. 2. Keep the progress JSON, the new audit JSON and the PDF together. Next month starts from them.
  175. ## Step 8: Recap in chat
  176. Keep it short:
  177. > **[domain]: [baseline date] → [today]**
  178. > AI Search Readiness Index **XX → YY** like for like (raw XX → YY including measurement change; range a–b,
  179. > [band]); [n checks improved, n slipped, on checks assessed both times].
  180. > Seen in AI, per engine: ChatGPT **x% → y%**, [engine] **x% → y%** …; [monthly: descriptive, the panel detects
  181. > only moves of about N points / quarterly: real change after the competitor control / within measurement noise /
  182. > platform drift].
  183. > **Fixed:** n · **Started:** n · **Still open:** n · **New:** n.
  184. > **Biggest win:** one sentence. **Biggest blocker:** one sentence. **Next three moves:** one line each.
  185. Then hand over the PDF. Offer, once, to set a reminder for next month's check; set it up only if the user says
  186. yes.
  187. ---
  188. ## Principles
  189. **Same ruler, every month.** Same scoring model (checks, pillars and weights), same depth, same prompts and
  190. run counts. If something has to
  191. change, change it once, note it in the method page, and keep it from then on.
  192. **Progress is what you saw, not what was planned.** A plan item marked done in a task list but not visible on
  193. the site is "in progress", with a note.
  194. **Name the blocker every month until it's gone.** If the same threshold fix (a stale index, Bing indexation)
  195. is open month after month, it leads the still-open page and the next steps.
  196. **Small numbers, honest framing.** A one-point move is noise until it repeats. Say so rather than celebrating
  197. it.
  198. **Keep the client's language.** Don't use terms the definitions page hasn't explained.
README.md · README · 115 lines
  1. # NoW Visibility Progress Check
  2. **A Claude skill from [The NoW of Work](https://nowofwork.com).** The monthly companion to the
  3. [NoW Search Visibility Audit](https://github.com/NoW-of-Work/now-search-visibility-audit).
  4. Once a site has had a NoW search visibility audit, this skill re-checks it and shows what actually moved, in the
  5. same designed, chart-led PDF format. It is built to answer one question honestly: did the work change anything,
  6. or is the difference just noise?
  7. ---
  8. ## What it does
  9. 1. **Finds the baseline.** It reads the original audit's data and score files. The baseline is always the
  10. original audit, not last month, so slow real gains aren't lost and false alarms don't pile up.
  11. 2. **Re-scores the site** with the audit skill, on the same scoring model, the same key pages and the same
  12. frozen panel of buyer questions. Checks are re-scored before the baseline scores are opened.
  13. 3. **Compares like for like.** The index is compared only over checks assessed in both audits. A check that
  14. merely became verifiable this time is reported as a measurement change, not progress.
  15. 4. **Sorts every earlier finding** into fixed, started, unchanged or new, and tracks each item in the 30/60/90 plan.
  16. 5. **Re-runs the same AI buyer prompts** per engine, against a control set of competitors, so a shift that hit
  17. everyone is called platform drift.
  18. 6. **Tests change honestly.** Monthly checks report mention rates for information. A formal "real change" or
  19. "within noise" verdict is given at the quarterly re-audit, against the original audit, per engine, with a
  20. Holm correction across engines.
  21. The report opens with a plain-words page explaining SEO, AEO and GEO and what each means for the business.
  22. **What a re-check can see.** At 20 prompts run once a day for 8 days, a re-check can only detect a change in mention
  23. rate of about 16 points or more. Precision comes from more prompts run on more separate days, not from more runs
  24. of the same prompt on the same days. Every report prints its own figure, so the client knows before buying
  25. monthly tracking.
  26. ---
  27. ## Using it
  28. > "Run the monthly progress check for example.com."
  29. >
  30. > "What's changed since the last audit?"
  31. >
  32. > "Show the progress since October."
  33. For a first audit with no baseline, use the
  34. [NoW Search Visibility Audit](https://github.com/NoW-of-Work/now-search-visibility-audit) instead.
  35. ---
  36. ## Installation
  37. **Claude desktop app or claude.ai**
  38. 1. Download `now-visibility-progress-check.zip` from the
  39. [latest release](https://github.com/NoW-of-Work/now-visibility-progress-check/releases/latest).
  40. 2. In Claude, open **Settings → Capabilities → Skills** and click **+**.
  41. 3. Upload the ZIP. Install the audit skill the same way; this skill uses it for the re-score.
  42. **Claude Code**
  43. ```bash
  44. git clone https://github.com/NoW-of-Work/now-visibility-progress-check.git ~/.claude/skills/now-visibility-progress-check
  45. ```
  46. ### Requirements
  47. - Web fetch and web search, and the NoW Search Visibility Audit skill (v7.0 or later).
  48. - Node.js 18 or later (no packages to install) and any headless Chromium for the PDF: Chrome, Edge, Chromium,
  49. Brave, or Python Playwright. Without one you get a print-ready HTML file.
  50. - Optional: OpenSEO, Google Analytics or Search Console connectors for sharper numbers.
  51. ---
  52. ## How it's built
  53. ```
  54. now-visibility-progress-check/
  55. ├── SKILL.md ← the workflow Claude follows
  56. ├── references/
  57. │ └── progress-method.md ← re-audit rules, statistics, the data file field by field
  58. ├── assets/report/
  59. │ ├── render-progress.mjs ← builds the progress report from its data file
  60. │ ├── scoring.mjs, charts.mjs, components.mjs, report.css ← shared with the audit skill
  61. │ ├── to-pdf.mjs ← prints the HTML with any headless Chromium
  62. │ └── example-progress-v7.json ← a complete worked example (fictional firm)
  63. ├── LICENSE
  64. └── README.md
  65. ```
  66. The shared files in `assets/report/` are byte-identical copies of the audit skill's. If you change one, change
  67. both.
  68. ---
  69. ## Version history
  70. **1.2.1**: Report layout fix. Long labels wrap instead of shrinking the whole PDF (shared `report.css`, same as audit 7.0.1). Method unchanged.
  71. **1.2.0**: First public release. Scoring model v7.0: like-for-like index on checks assessed in both audits,
  72. fixed original baseline, descriptive monthly checks and formal quarterly tests (per engine, cluster-robust
  73. errors, t reference, Holm correction, competitor drift), blind re-scoring, honest detectable change, refusals
  74. for mismatched models, tiers or data shapes.
  75. ---
  76. ## Credits and licence
  77. © 2026 The NoW of Work. Licensed under [Creative Commons Attribution 4.0 International (CC BY
  78. 4.0)](https://creativecommons.org/licenses/by/4.0/). You can use, share and adapt it, including commercially, as
  79. long as you credit The NoW of Work and link to the licence. See `LICENSE`.
  80. The licence covers the skill's text, not The NoW of Work's name or logo. "The NoW of Work" and "The Future is
  81. NoW" are marks of The NoW of Work. If you publish a modified version, remove the NoW branding from the report or
  82. make clear it isn't an official NoW of Work release.
  83. Questions or client work: [ai@nowofwork.com](mailto:ai@nowofwork.com)

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.