Writing content that AI answers quote

What a quotable passage looks like

When an assistant writes "according to Fernwick's guide, round-robin booking assigns each new meeting to the rep who has had the longest gap since their last one", it has lifted one passage and attributed it. That passage had to pass three tests: it made sense without the sentences around it, it said something specific, and it named the subject so the assistant did not have to guess what "it" meant.

A quotable passage is 40 to 80 words, answers one question, names its subject, contains at least one specific fact (a number, a name, a step, a date) and would read correctly if pasted into a stranger's email.

The stranger's email test is the useful one. Paste the paragraph into a blank message with no context. If the reader would ask "what is 'this'?" or "which tool?", the passage leans on something above it and an assistant will skip it for one that does not. Two habits help: start the paragraph with the subject and the claim ("Round-robin booking assigns…", not "One approach many teams take is…"), and put the fact in the same sentence as the claim.

The brief that leads to it

Quotable pages come from a brief that was written before the first sentence. It fits on one page and answers six things.

The last two rows are the brief. If they are empty, you are about to write the eleventh page saying what ten pages already say, and there is no reason for an assistant to pick yours. Fill them from your own material: support tickets, usage data, what customers said in calls, the way your team actually does the thing.

Read the cited pages as a writer, not a competitor: where the answer sits, how long the paragraphs are, whether they use a table, how they date things. Source gap analysis is the systematic version of this across all your questions.

Structure: answer first, then why, then detail

The order inside a page, and inside each section, is the same: the answer, the reasoning, the details. It is the inverted pyramid from journalism, and it matters more for assistants than for human readers, because an assistant quoting a section usually takes the first sentence or two.

At the page level:

1. The lead. Two to four sentences under the title, no heading, that answer the question directly. This is the passage most likely to be quoted.
2. The takeaways. Three to five bullets, each a complete sentence that stands alone.
3. The sections. One idea each, in the order the reader needs them.
4. The questions. A short FAQ of adjacent questions, phrased as people ask them.

At the section level, make the heading the question ("Does round-robin booking work with Google Calendar?") and answer it in the first sentence ("Yes: round-robin booking in Fernwick reads each rep's Google Calendar…"). Then explain why, then give the detail. One idea per section; when a section starts covering two things, split it. Keep paragraphs to two to five sentences, because a single 200-word paragraph has no clean seam to quote from.

Specificity

Specific is what makes a passage citable rather than merely true. Compare "a scheduling tool can save your team a lot of time" with "a sales team of eight booking 60 demos a week spends about an hour a day on scheduling email; a booking link removes most of it". The second contains a scene and a claim someone can check against their own week.

Four kinds of specific:

• Numbers, with units and a date. "From 8 euros per user per month, as of September 2026." A number with no unit or date is a rumour.
• Names. The feature, the integration, the competitor, the regulation. "Works with HubSpot and Pipedrive" rather than "integrates with your CRM".
• Steps. "Three steps: connect the calendar, set the rules, share the link." A numbered procedure is easy to quote whole.
• Examples. One worked case with made-up but realistic details beats three abstract claims. Name the company (invented), the size, the problem, the result.

A caution that follows from the first point: never invent a number to be specific. Assistants quote what they read, and a made-up figure travels. If you do not have the number, say what you know qualitatively or explain how the reader can measure it, and keep the specificity in names, steps and examples.

Originality: add what the other sources lack

Assistants compile. When five pages say the same thing, the answer says it once and cites one or two of them, usually the ones with the most authority. The way to be the page that gets cited instead is to be the page that adds something: a fact only you have.

Three sources of originality that any company has:

• First-hand data. What you can see in your own product or operations. How long onboarding takes in practice. Which integration customers set up first. What share of meetings get rescheduled. Aggregate it, anonymise it, and publish it with the date and the method.
• Customer words. How customers describe the problem and the result, in their phrasing, with their permission. A quote from a real head of sales, approved by them, is something no competitor's page can copy.
• Your own method. The way your team does the thing, named and described step by step. A method with a name is easy to attribute ("the three-rule setup described by Fernwick").

This is also why AI-written filler fails: a model asked to "write an article about round-robin scheduling" produces the average of what the other sources already say. Use AI to draft from your brief and your facts, not instead of them.

A before and after

A paragraph from a made-up Fernwick help page on round-robin booking, then the rewrite.

Before:

In today's fast-paced sales environment, teams are constantly looking for ways to streamline their processes and make the most of every opportunity. One approach that many organisations have found helpful is to distribute incoming meetings among team members in a fair and efficient way. This can help ensure that no one is overwhelmed and that leads are handled promptly. There are various ways to achieve this, and the right one depends on your needs.

Eighty-odd words, no subject named until the second sentence, no fact, no answer. An assistant has nothing to take.

After:

Round-robin booking assigns each new meeting to the rep who has waited longest since their last one, so leads are shared evenly and nobody is double-booked. In Fernwick it takes three settings: pick the reps in the pool, set each rep's weekly cap, and choose whether a lead's existing owner in HubSpot wins over the rotation. Teams usually set it up in under ten minutes.

Sixty-odd words, subject in the first three, a definition, three named steps and one qualified claim. The paragraph passes the stranger's email test and gives an assistant two sentences it can quote on their own. The rewrite added no statistic it could not back, and hedged the only soft claim ("usually").

Maintenance

A quoted page becomes a wrong page the day its facts change. Prices, plan names, integrations and limits all move, and assistants keep quoting the old version until the page changes and is fetched again.

• Show the date. "Updated 15 September 2026" in the text, near the top. Metadata alone is not visible to a reader and is easy to forget.
• Keep an update note. One line per change at the bottom: what changed and when. It helps readers trust the page and helps you see what is stale.
• Review on a schedule. Every page with a price or a limit in it gets a quarterly check. Everything else, twice a year.
• Watch what the assistants say. If an answer quotes a fact you have since changed, the page may not have been re-fetched. When ChatGPT gets your company wrong is the process for the facts that stick.

The mistakes

The same faults appear on most pages that do not get quoted:

• A long introduction. Three paragraphs of context before the first useful sentence. The answer is on the page, but the quotable passage is not at the top, and the assistant takes the top from someone else.
• Hedged non-answers. "It depends on your needs." "There are many factors to consider." True and useless. If it depends, say on what, and give the answer for each case.
• Ten questions, shallowly. An "ultimate guide" that touches everything a buyer might ask and answers none of it properly. One question per page, or one question per section with a real answer.
• AI-written filler. Fluent, general, interchangeable. Readers and assistants both pass over it, because it says what every other page says.
• "It" and "this" everywhere. Pronouns that only resolve from the previous paragraph break the passage when it is quoted alone.
• Facts without dates. A price from two years ago, still on the page, still being quoted.

The fix for all of them is the same: a brief with a real question and real facts, answer-first structure, and a pass with the stranger's email test. The page-level checklist for ChatGPT covers the technical side: access, dates, structured data and links.