Web Audit+

Open-source Chrome extension

Audit the page. Ship the fix.

Web Audit+ scores the tab you are on for Performance, Accessibility, Best Practices, SEO and a fifth category, Agentic, for how well the page serves AI agents. Every failing audit comes with a fix prompt written for your coding assistant. Nothing leaves your browser.

Add to Chrome See what it does
  • Free on the Chrome Web Store
  • No account
  • Runs locally
  • Chrome 125 or later
5scores on every run
22deterministic Agentic checks
4export formats: PDF, HTML, JSON, markdown
0servers, accounts or trackers

What you get

A report you can work from, not just read.

Failed audits come first and passed ones are folded away. Every finding quotes what it found on the page, says why it matters, and ends with the fix.

A fix prompt for every failing audit

Each prompt names the audit, quotes the evidence (elements, URLs, measured values), explains why it matters, states the fix and says what done looks like. Copy one audit, a whole category, or everything at once, and paste it into the coding assistant you already use.

### color-contrast: Background and foreground colors do not have a sufficient contrast ratio (Accessibility)

- Points at stake: +7 on the Accessibility score
- Evidence: 3 elements, including `p.muted` at `#pricing` (2.9:1, needs 4.5:1)
- Fix: raise the text color to at least 4.5:1 against its background
- Reference: https://dequeuniversity.com/rules/axe/4.10/color-contrast
- Done when: every text element meets 4.5:1 (3:1 for large text). Re-run Web Audit+ to confirm.

Fix next

Every fix ranked by the points it adds to the score, across all five categories. Start at the top.

1A5 raw-HTML content+9
2color-contrast+7
3meta-description+6

Compare runs

Audit the same URL again and the report shows what changed: fixed, new, still failing, and the score delta for each category.

Fixedimage-alt, link-text+9
Newunsized-images-3
SEO75 to 100+25

Exports that hold up

A print-ready PDF with selectable text, working links, bookmarks and tags for screen readers. The full run as JSON. A single-file HTML report that works offline. A markdown summary to paste.

PDFHTMLJSONMarkdown

Built for the keyboard

Search, move and copy without touching the mouse.

/
Search audits
j k
Next and previous audit
Enter
Open or close the audit
c
Copy its fix prompt
Esc
Leave the search

Private by default

Runs stay in your browser, newest 200. There is no account, no server and no telemetry. The only requests Web Audit+ makes go to the site you are auditing.

Read the privacy policy

Five scores

Lab-standard scores, plus the one nobody else gives you.

The first four categories are scored by the bundled audit engine, running on your tab with simulated throttling, device emulation and the standard metric weights and thresholds. The fifth is new.

Performance

Core Web Vitals and the metrics behind them, measured in a lab run on your own tab.

Accessibility

Contrast, names, roles, landmarks and keyboard access, checked against axe-core rules.

Best Practices

HTTPS, console errors, deprecated APIs, image sizing, security headers and the rest.

SEO

Titles, descriptions, canonicals, crawlability, structured data and legible text.

Agentic

Whether AI crawlers, search agents and assistants can find, read and trust the page.

The fifth score

Your next visitor is an agent. Agentic tells you what it sees.

Crawlers that build training sets, search agents that cite pages, and assistants that fetch a page because someone asked about it. Agentic runs 22 deterministic checks on how a page serves them, from robots.txt to a markdown twin served through content negotiation. No language model is involved: the same responses always produce the same score.

score = round(100 × earned / applicable)

Seven checks make the score: what AI agents need today. The other fifteen are suggestions, llms.txt, structured data and the markdown twin, reported with a fix each but outside the score, because the sites that serve agents best have not adopted them either. A check that cannot apply, such as HTTPS on localhost, leaves the denominator instead of dragging the score down. When the markdown twin is missing, the twelve checks that read it wait behind B1 and are reported as one line.

A1robots.txt exists and does not disallow known AI botsweight 9
A2llms.txt is present and validsuggestion
A3sitemap.xml is present and validweight 9
A4Schema.org JSON-LD is present and parsessuggestion
A5Real content is in the raw HTML, not only after JS runsweight 9
A6Time to first byte is fast enough for agent fetchesweight 8
A7No noai or noimageai tag is silently blocking AIweight 8
A8Served over HTTPS with HSTSweight 9
B1Markdown twin URL is reachable and returns 2xxsuggestion
B2Content-Type is exactly text/markdown; charset=utf-8suggestion
B3Markdown response body is non-empty and carries the pagesuggestion
B4X-Markdown-Tokens is present and a positive integersuggestion
B5X-Content-Type-Options: nosniff is presentsuggestion
B6X-Robots-Tag on the twin includes noindexsuggestion
B7Vary includes Accept, so CDNs cache the two formats separatelysuggestion
B8HTML response advertises the twin via Link rel="alternate"suggestion
B9Accept: text/markdown receives markdown, not HTMLsuggestion
B10A known AI bot User-Agent receives markdown without an Accept headersuggestion
B11An Accept header excluding HTML and markdown returns 406suggestion
B12X-AEO-Version advertises the spec version implementedsuggestion
B13The original HTML URL is still reachable and unaffectedweight 9
B14The served markdown still matches the live pagesuggestion

Status codes are never trusted on their own. A catch-all route that answers robots.txt with its HTML shell and a 200 is detected by reading the body, and scored as the missing file it is. How each check is scored.

How it works

Three steps. About a minute.

  1. Open the side panel

    On any page, click the Web Audit+ icon. Pick Desktop or Mobile and the categories you want, then press Run audit.

  2. Watch it run

    The Agentic score arrives in seconds. The audit engine then reloads the page in your tab with its own emulation and throttling and records a trace, exactly as a lab run should.

  3. Work the report

    Failed audits first. Search, filter, copy a fix prompt, export the PDF, and compare with the last run of the same URL.

While an audit runs, Chrome shows a bar saying that Web Audit+ started debugging this browser. That is Chrome telling you an extension is using the debugger, which the audit engine needs for emulation, throttling and tracing. It goes away when the run ends, and Cancel on it stops the run.

Privacy

Nothing leaves your browser. That is the whole architecture.

There is no backend to leak from. The extension talks to the site you audit and to nobody else.

0requests to anyone but the site you are auditing
200runs kept on your device, newest first, deletable any time
0accounts, analytics, telemetry or remote code

The full privacy policy, in plain language

FAQ

Questions, answered straight.

Is Web Audit+ free?

Yes. It is open source, there is no account, no plan and no usage limit. Install it and run as many audits as you like.

Why does Chrome say that Web Audit+ started debugging this browser?

That bar is how Chrome tells you an extension is using the debugger. The audit engine needs it to load the page with emulation and throttling and to record a trace, the same way a lab audit does. The bar goes away when the run ends, and pressing Cancel on it stops the run.

Does Web Audit+ send my pages anywhere?

No. The only requests it makes go to the site you are auditing. Runs are saved in your browser, there is no server, no account and no telemetry.

What does the Agentic score measure?

How well a page serves AI agents: whether crawlers are allowed in, whether a sitemap exists, whether the content is in the raw HTML, whether the page is fast and served over HTTPS. Those seven checks make the score. Another fifteen, llms.txt, structured data and a markdown twin served and negotiated correctly, are reported as suggestions with a fix each, outside the score: the sites that serve agents best have not adopted them either. It is 22 deterministic checks with fixed weights. No language model is involved.

Can I audit localhost or a staging site?

Yes. Everything runs from your browser, so anything your browser can open can be audited. The HTTPS check is marked not applicable on local hosts so it does not drag the score down.

Which browsers are supported?

Google Chrome 125 or later. The extension uses Chrome's side panel and debugger APIs.

Try it on this page

The report is the pitch.

Install the extension, come back to this page and press Run audit. This site was built to pass its own tests, and you can check.

Install Web Audit+ Read the source