ChatGPT Search: What SGE-SCORE Audits and How the Master-Fix Closes Gaps

How ChatGPT Search selects content

ChatGPT Search has been rolled out to Plus, Team and Enterprise users since November 2024 and is available to all ChatGPT Free accounts since early 2025. Source selection runs in two stages: candidates are first retrieved from an index built by the OAI-SearchBot crawler and the training crawler GPTBot. An internal reranking model then scores each candidate page on relevance, freshness, source quality and readability for the language model.

Unlike classic Google crawlers, OpenAI puts particular weight on machine-readable main content: when the main content only appears after JavaScript hydration or sits behind cookie banners, a page falls behind. OpenAI's official bot documentation explicitly states that OAI-SearchBot renders JavaScript only to a limited extent.

"Citation in ChatGPT does not go to the loudest page — it goes to the clearest one."

Whoever wants to be cited in ChatGPT therefore has to deliver three things at once: a clean crawl path (robots, llms.txt, sitemap), instantly parseable main content, and a clear trust signal covering author, sources and freshness. That's exactly what our audit checks — the SGE-GEO knowledge hub ties every signal back to a patent or official source.

→ Next step: Run our free demo analysis to score your domain on all 57 criteria.

The 7 most important signals for ChatGPT visibility

1. Crawl discoverability: OAI-SearchBot and GPTBot must be allowed to fetch important paths — checked in robots.txt.
2. llms.txt: A short, curated map of your most important content under /.well-known/llms.txt — a "human-readable sitemap.xml for LLMs".
3. Main content without JS rendering: Headings, lists and core copy must appear in the initial HTML response.
4. Direct-answer block: 40–60 words of clear answer at the top of the article — Princeton GEO paper, 2023.
5. Source transparency: External links with rel="noopener", clear date stamps, author bio with verifiable sameAs.
6. Structured data: At minimum Article + FAQPage + BreadcrumbList; optionally HowTo.
7. Chunk structure: Sections ≤ 250 words with their own H2/H3 — so the model can reference cleanly.

→ Next step: Compare your current article against this list — anything missing goes into the Master-Fix prompt.

What our audit specifically checks for ChatGPT

SGE-SCORE evaluates 57 patent- and best-practice checkpoints (max. 114 points). For ChatGPT visibility, these 12 are particularly relevant:

For each checkpoint we output a score (0 / 1 / 2), a concrete justification and a patent or source reference. More details in the documentation and in the entry building llms.txt correctly.

→ Next step: Read the related article Brand Mention Index & authority for SGE success — it explains the trust signal that ChatGPT also relies on.

Before / after: what the Master-Fix actually produces

The Master-Fix takes the audit result and generates a complete patch set in roughly 60 seconds: HTML patches, schema JSON, llms.txt proposal and author bio. Example:

Before (typical marketing page)

<article>
  <h1>Our solution</h1>
  <div id="root"></div>  <!-- content arrives only after JS -->
</article>

After (Master-Fix output)

<article itemscope itemtype="https://schema.org/Article">
  <header>
    <h1 itemprop="headline">Our solution for AI visibility</h1>
    <p class="lede">In 60 seconds we show which of the 57 checkpoints
    your domain passed — and where the Master-Fix steps in.</p>
    <p class="meta">
      By <span itemprop="author">Pascal Schildknecht</span> ·
      <time itemprop="datePublished" datetime="2026-04-30">30 Apr 2026</time>
    </p>
  </header>
  <section id="key-takeaway">...</section>
</article>

The Master-Fix additionally ships an embedded Article, FAQPage and BreadcrumbList schema as well as an llms.txt block you can drop in 1:1 at /.well-known/llms.txt.

→ Next step: If you have not unlocked the Master-Fix yet, our €9 promo is currently live.

llms.txt: the curated map for ChatGPT & co.

/.well-known/llms.txt is not an official OpenAI standard, but it has become a de-facto convention (see llmstxt.org). The file tells an LLM crawler: "here is my most important content, in readable order".

# SGE-SCORE — AI-readiness audit for landing pages

> SGE-SCORE audits websites against 57 patent-based checkpoints for Google AI Overviews,
> ChatGPT, Gemini and Claude — and ships actionable patches via the Master-Fix.

## Most important content
- [What we audit](https://sge-score.com/en): patent knowledge, 4 sectors
- [Master-Fix](https://sge-score.com/start-demo): €9 promo, output in 60 s
- [Documentation](https://sge-score.com/en): methodology, API, integrations

## Optional
- [Knowledge base](https://sge-score.com/en): ~1,520 entries

A valid llms.txt earns 2 audit points and is, based on our internal tests, one of the most common quick wins — many domains simply do not have it.

→ Next step: The Master-Fix automatically delivers a ready-to-use llms.txt proposal in your clipboard.

The 5 most common mistakes in ChatGPT optimisation

• OAI-SearchBot blocked: the default robots.txt permits Googlebot but blocks unknown crawlers via generic rules.
• Pure SPAs without SSR: main content is built only client-side — the crawler sees nothing.
• Cookie banner hides main content: even if the HTML is there, the crawler often only parses the visible area.
• Missing author bio: without a Person schema there is no trust signal — and no Knowledge-Graph link.
• Stale date: ChatGPT prefers sources with a current dateModified.

All five mistakes are explicitly flagged in our audit — see also the "Trust" section in the SGE-GEO knowledge hub.

Frequently asked questions (FAQ)