CiteScore

Checklist

ChatGPT citation checklist for SaaS homepages

Twelve on-page checks that make a SaaS homepage easier for ChatGPT, Perplexity, and Google AI Overviews to quote — without pretending anyone logged into ChatGPT for you.

14 September 2026 · 9 min read

This is a homepage checklist, not a promise that ChatGPT will name you on Tuesday. CiteScore never logs into ChatGPT. We look at whether a model could extract a definition, an entity, and a fact from the public HTML — then we say so in the report. Use the list below before you pay anyone, including us.

If you want the longer definition of the category, start with what GEO is. If you want the scored version of this list on your URL, that is the $39 GEO audit — also described as a ChatGPT citation check for the search that asks for one.

  1. 01Lead with a definition, not a slogan

    If the H1 and the first 80 words cannot finish the sentence “This product is a … that helps …,” a model has nothing safe to lift. Slogans are fine in the nav. They are a poor citation.

  2. 02Repeat the product name as an entity

    Same spelling in the title, H1, and meta description. Then say it again in Organization or SoftwareApplication JSON-LD. Models attach claims to names; a nickname-only hero is easy to skip.

  3. 03Put a real Q→A on the homepage

    Two or three question headings (“What is …?”, “Who is it for?”, “How is this different from …?”) with short answers beat a wall of feature tiles. FAQPage markup on those pairs helps extraction.

  4. 04Offer one dated, named fact

    A number, a year, a method. “Used by 12,000 teams” is weak if it never changes and has no source. “In our 2026 survey of 180 agencies, 41% …” is something a model can quote with a straight face.

  5. 05Name a human on an About or authors surface

    Unattributed marketing pages are easy to treat as brochureware. A Person schema, a byline, or a short About with credentials raises source-worthiness. You do not need a thought-leadership blog to do this.

  6. 06Show that the page is alive

    A visible last-reviewed date, a changelog, or a /blog with a current year. Stale pages lose to last year’s roundup posts even when your product is better.

  7. 07Do not block the bots you want citations from

    Check robots.txt for GPTBot, PerplexityBot, and Google-Extended. A leftover “Disallow: /” from a staging deploy is an own-goal. If you block them on purpose, say so — and read the crawler dimension with that policy in mind. The audit’s exact rules are on the AI crawlers page, including the user-agents it does not parse.

  8. 08Add llms.txt if you have a canonical explainer

    llms.txt is a small Markdown map of the pages you want a model to read. It is not magic. CiteScore only checks that /llms.txt returned a non-empty body, and counts that as a positive crawler-access signal. The llms.txt page is the exact check, including what a missing file does to the score.

  9. 09Ship a comparison or alternatives URL

    Buyers ask “X vs Y.” If you have no /compare or /alternatives page, the model will use a third-party listicle. Write the honest version yourself. We did that for Otterly; you should do it for your actual rivals.

  10. 10Keep key same-host pages fetchable

    Homepage, /about, /pricing, /faq, and one compare URL. A crawler that only sees a JS shell and a cookie wall will score the site as thin, because that is what it got.

  11. 11Give the page a canonical and sane Open Graph

    Canonical URLs stop the same product being split across www, trailing slashes, and campaign params. og:title / og:description are not vanity — they are another labeled copy of the entity.

  12. 12Write the answer you want quoted

    If you want ChatGPT to say “Northbound is a customer-interview repository for product teams,” that sentence has to exist on the page, early, in prose a human would also accept. Do not hide it in a screenshot.

What this checklist is not

It is not a live citation rank. Prompt wording, user location, and whether browsing is on will move any single ChatGPT answer. The AI Overviews and Perplexity pages describe the other two rows in the same PDF. Enterprise monitors such as Otterly or Profound exist to sample prompts over time. That is a different product — and a different price.

It is also not an excuse to keyword-stuff “ChatGPT” into every H2. Answer engines are not impressed by that, and neither are buyers. Write the page for a skeptical founder. If a model can quote it, that is a side effect of being clear.

See the Northbound sample report for how these checks show up as dimension scores and a ten-item fix list. The company is fictional. The format is the one you buy.