Do I always need a Firecrawl key?

No, only for fetching URLs. The 'paste content directly' expander reviews pasted markdown or text with just your Anthropic or OpenAI key, so you can skip Firecrawl entirely if you already have the content.

Why is some of my page missing from the review?

The Firecrawl scrape uses onlyMainContent plus an includeTags whitelist of h1-h6, p, li, table and blockquote. Text in other elements (divs, spans, accordions rendered without those tags) and anything Firecrawl classes as navigation or ads never reaches the model, so it cannot be reviewed. Paste the content directly if coverage matters.

Can the annotated output get cut off?

Yes. The review call is capped at 8,000 output tokens, and because the output repeats your full content with annotations inserted, long pages can hit the cap and lose the end of the review, including the summary section. Review very long pages in parts via paste mode.

What are the limits and pacing in bulk mode?

Bulk runs are capped at 50 URLs per run (slider, default 10) with a fixed 2 second pause between URLs. Each URL costs one Firecrawl scrape plus one LLM call over the full fetched content, so token spend scales directly with page length.

Are the annotations machine-readable?

They are HTML comments inserted into the markdown (ISSUE, MISSING, STRENGTH, IMPROVE, SEO), which means they are invisible when the markdown is rendered but easy to grep or parse from the downloaded file.

Content Reviewer (LLM): free Content tool

#about

Firecrawl API (/v1/scrape) with 30s timeout, main content extraction, ad removal.

Claude Sonnet 4.5/4, Haiku 4.5 or GPT-4o/4o-mini.

Four review presets: SEO Quality, Technical Accuracy, UX, Conversion.

HTML comment annotations: ISSUE, MISSING, STRENGTH, IMPROVE, SEO.

GPT-4.1 is also selectable.

Streamlit appRequires an API key

Platform: Browser-based (no installation required)
Input: Firecrawl API keyAnthropic or OpenAI API keyURL to review or content paste
Output: Annotated markdown with ISSUE/MISSING/STRENGTH/IMPROVE/SEO comments. Summary section with quick wins.
Category: Content

Use cases

Content quality audits
Pre-publish content review
Competitor content analysis
Training content writers

#features

Firecrawl scraping: main content only, whitelist of heading/paragraph/list/table tags
Paste mode needs no Firecrawl key
Four review presets + custom focus
Claude Sonnet 4.5/4, Haiku 4.5 or GPT-4o/4o-mini/4.1
Inline HTML comment annotations with priority summary
Bulk mode: up to 50 URLs per run, 2s pause between URLs

#how-to-use

Enter AI API key (plus Firecrawl key for URL fetching)
Select review focus (SEO/Accuracy/UX/Conversion/Custom)
Provide URL or paste content directly
Review annotated output with inline comments
Download markdown with priority summary

#faq

Do I always need a Firecrawl key?: No, only for fetching URLs. The 'paste content directly' expander reviews pasted markdown or text with just your Anthropic or OpenAI key, so you can skip Firecrawl entirely if you already have the content.
Why is some of my page missing from the review?: The Firecrawl scrape uses onlyMainContent plus an includeTags whitelist of h1-h6, p, li, table and blockquote. Text in other elements (divs, spans, accordions rendered without those tags) and anything Firecrawl classes as navigation or ads never reaches the model, so it cannot be reviewed. Paste the content directly if coverage matters.
Can the annotated output get cut off?: Yes. The review call is capped at 8,000 output tokens, and because the output repeats your full content with annotations inserted, long pages can hit the cap and lose the end of the review, including the summary section. Review very long pages in parts via paste mode.
What are the limits and pacing in bulk mode?: Bulk runs are capped at 50 URLs per run (slider, default 10) with a fixed 2 second pause between URLs. Each URL costs one Firecrawl scrape plus one LLM call over the full fetched content, so token spend scales directly with page length.
Are the annotations machine-readable?: They are HTML comments inserted into the markdown (ISSUE, MISSING, STRENGTH, IMPROVE, SEO), which means they are invisible when the markdown is rendered but easy to grep or parse from the downloaded file.

/contact

Want me to run this for you? I run this tool as a managed service, or build something custom around your data. You get the insights without touching the code.

Send me a message →