KorraNet Creative
How every number in the report is produced, and where the measurement stops.
A measurement without a stated boundary is an opinion. What follows is the boundary.
| Section | The question |
|---|---|
| Crawler access | Can GPTBot, PerplexityBot, ClaudeBot and Google-Extended reach this page |
| What the crawler receives | Does the raw HTML carry the content, or is the page assembled by JavaScript |
| Headings | One H1 or several. Duplicates, skipped levels, questions |
| Title and meta | Length, canonical, lang, Open Graph |
| Structured data | Which schema types are present, missing or duplicated |
| Extractability | Is there a single passage an answer engine can lift whole |
| Assets and links | Alt text, internal links, empty anchors |
A robots.txt states a policy. This checks what
the server actually does.
Every audit makes a second request to the same page carrying OpenAI's own user agent.
A site can publish a welcoming robots.txt and still return
403 to GPTBot at the edge, through bot rules nobody remembers turning on.
That gap is invisible to any tool that only reads robots.txt.
Every limit below is a real one.
| Limit | What happens |
|---|---|
| Pages over 8 MB | Read in part. The report says so at the top and tells you not to send it. |
| Rendered state | Not measured. Accordions, tabs and anything JavaScript reveals need the console version on a laptop. |
| The GPTBot line | Records what the server returns to that user agent. It does not prove how OpenAI uses the page. |
llms.txt | Checked because it is free to check, not because there is evidence any crawler reads it. Ahrefs found 97% of sites with one were never asked for it. |
| Which findings matter | Not measured. That is a judgement call and it is not in this app. |
Every report carries the tool version that produced it. When a version is found to be wrong it is recorded here, and anyone who received an affected report is told.
| Version | Fixed | Effect on earlier reports |
|---|---|---|
| 1.4 5 Nov 2026 |
Paragraph count reported the display cap of 4 instead of the real count. Cookie and consent text was counted as a candidate entity statement. | Any count of 4 before 1.4 means "4 or more", and may include consent boilerplate. |
| 1.3 5 Nov 2026 |
The fetcher cut page bodies at 2,000,000 characters while still reporting the true size, so a large page was parsed in part and reported as whole. | Reports on pages over 2 MB before 1.3 are wrong. A 3.3 MB page was reported as having 0 headings and 175 words. It has 26 headings and 1,650 words. |
| 1.2 5 Nov 2026 |
Sitemap index files were counted as single-URL sitemaps. Headings in the header, nav and footer were counted against the page. Only <p> elements were treated as paragraphs. |
A healthy sitemap index could be flagged, template headings could appear as page faults, and pages using <div> for body text could be reported as having no quotable passage. |
| 1.1 4 Nov 2026 |
<script> and <style> content was counted as body text. |
Word counts and the numeric share of body text are wrong, and the thin-page check was suppressed. |
The audit runs in your browser. The fetcher performs GET
requests and keeps nothing: no log, no database, no analytics. The URLs you
have audited are stored in your own browser and go nowhere else. Private and
local network addresses are refused.