How to Write Pages ChatGPT Can Quote

Before writing this guide we ran Leadee's ChatGPT Citation Score — the tool embedded above — on 8 live Netbase-owned pages (all part of Netbase JSC, the company behind Leadee): the weakest page of netbasejsc.com's content audit (its web application development service page) and the homepages of the 7 Netbase sites Leadee tracks. The scorer splits a page into passages and checks four signals in each: does it open with a direct statement, does it contain a concrete number, does it attribute a claim to someone, and does it name a source.

The pattern was the same on every page. Almost every passage opened with a plain statement — these are professionally written sites. What was missing were the parts a model needs to quote a claim with confidence: numbers appeared in about one passage in 18, and a named source in about one in 40. The average passage score ranged from 0.22 to 0.30 on a 0-1 scale.

Two limits: the scorer reads every paragraph and list item, so menu blocks are counted as passages too, which pulls the averages down a little; and its number detector only recognises some shapes (percentages, prices, and counts such as "500 customers" or "12 years"), so "3 to 30 people" scores as no number. Treat the shares as approximate. They describe the pages, not whether ChatGPT cited them.

Start with a page that should be the answer to a question buyers actually ask an AI assistant: a service page, a pricing page, a comparison. Write the question down in the buyer's words, for example "how long does a custom web app take to build?". If you do not know which questions matter, the free ChatGPT Prompt Finder lists likely ones for your industry.

One page, one main question. A page that tries to answer ten questions usually answers none of them in a passage a model can lift out cleanly. Where this guide fits among the other ChatGPT steps is shown on the ChatGPT SEO overview; why the answer engine needs the page to answer clearly is covered in how AI chooses suppliers.

Run the page through the ChatGPT Citation Score and sort the passages by score. Ignore navigation and footer blocks. Mark the three to five passages that actually try to answer your buyer question — those are the ones worth rewriting, whatever their score.

Check before you edit that ChatGPT's search crawler can read the page at all. If OAI-SearchBot is blocked, no rewrite will help; what each OpenAI bot does explains which rule to look for.

Rewrite each marked passage so its first sentence answers the question on its own. Not "Every project is different", but "A new web application usually takes three to six months from discovery to launch" — if that is true for you. Lead-ins such as "Welcome to", "In this article" or a rhetorical question score as no direct answer.

Name the subject in that sentence. "We" and "our team" mean nothing once a passage is quoted out of context; "Netbase JSC" or your product name does.

Most pages already contain the numbers buyers want — team sizes, lead times, prices, years in business, customer counts — but in a different paragraph from the claim they support. Move the number next to the claim. Use a figure you can stand behind; never add one to please a scorer.

If you do not have the number, leave the passage without it and note what is missing. The Citation Score's rewrite pattern marks such gaps as "needs input" instead of inventing a figure, and so should you.

A claim a model can attribute is easier to cite than an unattributed one. Add "according to" with a real source — your own case study, a published report, a standard — and link it. For your own results, link the page where the full case is told.

Our before-and-after check used only facts already on netbasejsc.com, which is Netbase-owned (part of Netbase JSC). The original passage, "Within six months 4over4 reported 82% higher e-commerce revenue and a 48% higher conversion rate.", scored 0.50: it had a direct statement and numbers, but no subject, attribution or source. Our rewrite — "Netbase JSC reworked checkout, search and navigation for 4over4, a US online printing store. According to the 4over4 case study, 4over4 reported 82% higher e-commerce revenue and a 48% higher conversion rate within six months (source: netbasejsc.com/work/4over4-ecommerce-conversion)." — scored 1.00. That rewrite was made for this test and is not published on that Netbase-owned site.

Read each rewritten passage as if it were the only thing on the page. Remove references such as "as mentioned above" or "see the table below", spell out abbreviations once, and keep one idea per paragraph. Keep each passage short enough to quote whole — a few sentences, not a wall of text.

Where a set of facts is easier to read as a table (sizes, prices, lead times), use a real HTML table, not an image. Our schema markup guide covers marking the same facts up for machines.

Score the edited text again. Before the page is live, paste it into the general Content Score tool, which runs the same scorer and accepts pasted text; once it is live, re-run the ChatGPT Citation Score on the URL. The marked passages should now show all four signals where you had the facts to support them. Add a visible "updated" date to the page so readers and engines can see the facts are current, then publish.

A higher Citation Score means the passage now carries the signals answer engines tend to quote. It does not mean ChatGPT has quoted it. To know that, ask your buyer question repeatedly over several weeks and count how often your page is cited — one answer is one sample, and the same question can cite different pages each time.

Leadee's tracker does this through OpenAI's official API and reports a citation rate with a confidence interval; what ChatGPT citations are explains how the count works and why small samples mislead. Give it at least a few weeks after the page is re-crawled before judging.