Last updated: 2026-11-01

7 Mistakes That Tank Your ChatGPT Visibility Score

Assistants lift clear, self-contained statements from pages they can read and trust. Hide, bury, or muddy those statements and you lose the citation, even when the content is good. Here are the seven mistakes I see most often, plus a fix list that takes about ten minutes per page.

Mistake 1: hiding answers in images and popups

An assistant reading your page works from the text it can extract. Text baked into an image is invisible to it, and content that only appears after a click may never be seen (behavior checked 2026-11-01). If your pricing or your how-it-works steps live inside a hero image or a modal, assume the assistant never saw them. Keep the core answer as plain HTML text.

Mistake 2: writing for keywords instead of questions

Traditional SEO rewards repeating a keyword in headings. Assistants respond to full questions. A page titled "Pricing" that lists numbers without saying who the product is for gives the assistant little to quote. Rewrite key sections as direct answers to the questions your buyers actually ask, and use those questions as your subheadings.

Mistake 3: burying the answer below the fold

Long introductions push the substance far down the page. When an assistant extracts a passage, it prefers content that states the answer early and plainly (behavior checked 2026-11-01). Put a two to three sentence direct answer right after the heading, then expand below it. Human readers benefit too.

Mistake 4: no dates, no sources, no verifiable facts

Assistants weigh recency and traceability when deciding what to quote (behavior checked 2026-11-01). A page with no publication date and no named sources reads like an opinion floating in time. Add a visible last-updated date and cite key claims with outbound links. Our own tools state clearly that they are heuristic self-assessments that do not query any AI live. Copy that kind of honesty.

Mistake 5: skipping structured data entirely

Schema markup like FAQPage or Article gives machines an explicit map of your page. It cannot force any assistant to cite you, but it removes ambiguity about which text is a question, which is an answer, and who wrote it. Generating valid JSON-LD takes minutes with a free generator; validate it with a rich-results tester before you ship.

Mistake 6: publishing content that needs JavaScript to render

If your article body loads through client-side JavaScript, some readers, including crawlers and assistants with limited browsing, may see an empty shell instead of your text (behavior checked 2026-11-01). Fetch the page's raw HTML: if the article text is missing there, server-render the content or provide a static version.

Mistake 7: skipping machine-readable summaries

An llms.txt file at your domain root listing your key pages and what each covers is becoming a common convention. Keep it honest: describe what the pages actually contain. Refresh it whenever you add a new section, the same way you would refresh a sitemap.

The 10-minute fix list

Work through this list one page at a time:

1. Pull your main answer out of images and popups into plain text.
2. Add a two to three sentence direct answer under the main heading.
3. Rewrite two subheadings as the exact questions your buyers ask.
4. Stamp the page with a visible last-updated date and link key claims to sources.
5. Add one FAQPage JSON-LD block covering three to five real questions.
6. Check the page with JavaScript disabled and confirm the text is still there.
7. Add or refresh llms.txt so your key pages stay discoverable.

Then run the page through a visibility checker, write down the score, and repeat monthly.