SEO Fix Agent

Every SEO issue, engraved into a fix your agent can ship

36 checks across technical, meta, content, links, social and AI search — each one weighted, so the score moves by what a problem actually costs you. Everything that fails compiles into one severity-ordered prompt.

Runs now, no account. The fixes and the compiled prompt come with a free one — 50 credits, no card.

36 checks · 29 weighted · example.com

74/100C

  • AI answer engines can fetch this page: passed, weight 3 of 45.
  • HTTPS enabled: passed, weight 3 of 45.
  • Page is indexable: passed, weight 3 of 45.
  • Site not blocked by robots.txt: passed, weight 3 of 45.
  • HTTP status code: passed, weight 3 of 45.
  • Title tag: warning, weight 3 of 45.
  • Canonical URL: warning, weight 2 of 45.
  • H1 heading: passed, weight 2 of 45.
  • Meta description: failed, weight 2 of 45.
  • Server response time: passed, weight 2 of 45.
  • Structured data (JSON-LD): warning, weight 2 of 45.
  • Mobile viewport configured: passed, weight 2 of 45.
  • Content length: failed, weight 2 of 45.
  • Character encoding declared: warning, weight 1 of 45.
  • Text compression: passed, weight 1 of 45.
  • Descriptive anchor text: passed, weight 1 of 45.
  • Heading hierarchy: passed, weight 1 of 45.
  • Internal linking: failed, weight 1 of 45.
  • HTML lang attribute: passed, weight 1 of 45.
  • Open Graph tags: failed, weight 1 of 45.
  • HTML document size: passed, weight 1 of 45.
  • robots.txt present: warning, weight 1 of 45.
  • XML sitemap reachable: warning, weight 1 of 45.
  • Answer-first content: passed, weight 0.5 of 45.
  • Authorship / E-E-A-T signals: warning, weight 0.5 of 45.
  • Favicon present: passed, weight 0.5 of 45.
  • Image lazy loading: passed, weight 0.5 of 45.
  • Title keyword stuffing: passed, weight 0.5 of 45.
  • Twitter/X card: warning, weight 0.5 of 45.

Every segment is one check; its width is that check’s real weight. Filled area is the score. A missing title costs 3; a missing hreflang costs nothing at all.

Reported · not scoredhreflangimg-altexternal-linksfaq-schemaai-training-accessllms-txtcontent-freshnessRun and reported, never scored — blocking AI training crawlers is a licensing choice, not a defect.

Nothing on this page is a placeholder

There are no testimonials here, because there is nobody to quote yet, and inventing them is not something we will do. What we have instead is a product that can be checked — so check it.

The bar above is a real audit of example.com, recorded 2 September 2026, with the engine’s own weights. Run the same audit on this site yourself — it costs one credit, and you start with fifty.

Our own score
100/100 · grade A
Lighthouse
100SEO100Accessibility100Best practices
Measured
21 August 2026, and re-measured in CI on every release. If a number here drifts from what the engine reports, the build fails.

Nine tools

0 mocked · 0 coming soon

Three demo-data tools — rank tracking, backlinks and AI mentions — were removed rather than shipped with invented numbers behind them. They are not listed above because they do not exist.

Audit, compile, ship, re-run

  1. 1

    Point it at a URL

    The engine fetches your page the way a crawler does — plus robots.txt and your sitemap. Nothing to install, no OAuth, no access to your repo or your analytics.

  2. 2

    Read one score and a ranked list

    All 36 checks run, each one weighted. What comes back is a 0–100 score you can measure off the bar above, and every finding ordered by what it actually costs you.

  3. 3

    Compile the findings into a prompt

    Prompt Studio turns the result into one severity-ordered instruction set for Claude Code, Cursor or ChatGPT — the exact change spelled out, highest impact first.

  4. 4

    Ship it with your agent

    Paste, review the diff, deploy. The fix happens in your editor, in your repo, under your review — not in a dashboard you have to translate back into code.

  5. 5

    Re-run and watch it move

    Audit again and compare. Schedule it, and we email you only when something regresses — silence from us is a real all-clear.

36 checks, and what each one is worth

Weight is what separates this from a checklist. Three checks are reported and never scored at all — blocking an AI training crawler is a licensing decision, not a defect, and it would be dishonest to charge you points for making it.

Category 01/06

Technical

12 checks

  • w3HTTPS enabled
  • w3Page is indexable
  • w3Site not blocked by robots.txt
  • w3HTTP status code
  • w2Server response time
  • w2Mobile viewport configured
  • w1Character encoding declared
  • w1Text compression
  • w1HTML document size
  • w1robots.txt present
  • w1XML sitemap reachable
  • w0.5Favicon present

Category 02/06

Meta

6 checks

  • w3Title tag
  • w2Canonical URL
  • w2Meta description
  • w1HTML lang attribute
  • w0.5Title keyword stuffing
  • —Hreflang annotations

Category 03/06

Content

5 checks

  • w2H1 heading
  • w2Content length
  • w1Heading hierarchy
  • w1Image alt text
  • w0.5Image lazy loading

Category 04/06

Links

3 checks

  • w1Descriptive anchor text
  • w1Internal linking
  • —Outbound links

Category 05/06

Social

2 checks

  • w1Open Graph tags
  • w0.5Twitter/X card

Category 06/06

GEO / AI search

8 checks

  • w3AI answer engines can fetch this page
  • w2Structured data (JSON-LD)
  • w1llms.txt present
  • w0.5Answer-first content
  • w0.5Authorship / E-E-A-T signals
  • w0.5FAQ / Q&A schema
  • —AI training crawler stance
  • —Content freshness signals

The report is not the deliverable

Every finding becomes an instruction, and the instructions arrive in weight order — so the first thing your agent fixes is the thing costing you most. This is the audit from the top of the page, compiled.

No PDF, no ticket queue, no “schedule a call with an SEO consultant”. One block of text your coding agent can act on.

Compile your first prompt

Compiled · weight order

# Fix prompt — example.com
# 12 findings, ordered by weight. Highest impact first.

1. [w3] title
   Write a unique 30–60 character title with the primary keyword near the front.

2. [w2] canonical
   Add a self-referencing canonical link to prevent duplicate-content dilution.

3. [w2] meta-description
   Write a 70–160 character description with the target keyword and a clear value proposition.

+ 9 more, ranked the same way — down to w0.5 twitter-card.

One credit per run. That is the whole model

No seats, no feature gates, no tier that unlocks the checks you actually needed. Spend from the meter and see where fifty gets you.

Your balance on a free account

50of 50 runs left

Spend some. Every run costs one credit; the rest of the product costs nothing.

Free
50runs$05 on signup, 45 when you verify your email. No card.Create account
Pro
1,000runs$10every month. Cancel any time.Upgrade
Pro annual
12,000runs$96a year, granted up front. 20% off.Upgrade
Agency
6,000runs$49every month. Teams, client workspaces, branded PDFs on every plan.Upgrade
Top-up
1,000runs$10once. No subscription.Buy credits

Prices in USD. Tax is calculated at checkout where it applies.

Questions people actually ask

What do I actually get for free?

A full audit of any page from the homepage, without an account — every finding, scored. Then 50 credits when you create an account (5 immediately, 45 once you verify your email), no card required. A site audit costs 1 credit and a keyword search costs 1 credit — so that is 50 runs before you decide whether this is worth paying for. Prompt Studio, shared reports, projects, saved lists and CSV export do not cost credits at all.

Do you need access to my site, repo or analytics?

No. There is nothing to install and no OAuth to grant. The audit engine fetches your public URL, robots.txt and sitemap over plain HTTP, exactly the way a search crawler does.

How is the score weighted?

Every check carries a weight, and the score is the weighted proportion that passed — a warning earns exactly half its weight, a failure earns none. That is the bar at the top of this page: each segment's width is one check's real weight, and the filled area is the 74. A missing title costs three; a missing hreflang costs nothing, because we do not score it at all.

How is this different from Ahrefs or Semrush?

Those tools end at the report. SEO Fix Agent ends at the fix: every finding compiles into a severity-ordered prompt for Claude Code, Cursor or ChatGPT, so the distance from “issue found” to “issue shipped” is one paste and one review.

Why does the Keyword Explorer not show search volume?

Because we do not have it. The keywords are real — they come from live Google autosuggest expansion, and the intent classification runs on the actual phrases. What we show next to each one is its position in Google's own suggestion list, which Google orders by popularity: a real demand signal, but not a volume. We would rather show you a number we can stand behind than invent a licensed-clickstream figure we do not license.

What does GEO mean?

Generative Engine Optimization — making your pages citable by AI answer engines. The audit checks JSON-LD structured data, FAQ schema, answer-first paragraph structure, E-E-A-T signals, llms.txt, and whether your robots.txt is blocking the retrieval bots that feed AI answers.

Do credits expire, and can I cancel?

Credits never expire. Pro is a normal monthly Stripe subscription you can cancel any time from the billing page, and one-off top-ups need no subscription at all.

Can I get my data out?

Yes. Audits and keyword lists export to CSV, and the full audit result is stored as JSON you can pull from the API. Nothing is locked in.

From the blog

All writing

Fifty runs, no card, nothing to install

Start freeRead the writing first