By KYN AI Advisory Team — AI implementation specialists, Singapore
An answer-first article opens with the direct answer in two to four sentences, so an assistant or a hurried reader can quote the opening without reading the rest. The remaining sections justify the answer: question-style headings, specific facts with sources, a table where a comparison is real, and a short FAQ. No format ensures a citation. This one makes your page easy to lift accurately, which is the condition for being used.
What does answer-first mean?
It means the first paragraph answers the title's question and does not introduce it. Compare two openings for "How long does bank reconciliation take to automate?"
Weak: "Bank reconciliation is a critical part of any finance function. In this article we will explore the options available to growing businesses."
Strong: "A bank reconciliation can be automated in weeks if the bank feed is clean and the matching rules are written down. Most of the time goes into the rules, not the software."
The second can be quoted as it stands. It states a claim, a condition and the cause of delay. Everything after it supports and qualifies that claim.
Why does it matter for AI assistants?
Assistants retrieve passages and compose an answer from them. A passage that states the answer in its first lines is easier to retrieve and less likely to be quoted wrongly.
Be careful about what is proven. Google states in its guidance on AI features that no special optimisation or markup is required to appear in AI Overviews, and that a page must be indexed and eligible to appear in Search with a snippet. The rest, such as how passages are chosen, is not fully published. Answer-first writing is a sound practice because it also serves human readers who scan, not because a vendor has promised a result.
What is the structure?
Use the same skeleton on every article.
- Answer paragraph. Two to four sentences. Include the key condition or number.
- Four to seven H2 sections. Each opens with its own one-sentence answer.
- A comparison table where you are comparing real options on real attributes.
- A short section on who you are and how you fit. Factual, no sales language.
- Three FAQ questions, each answered in two to four sentences.
- A check date. "Checked 30 October 2026." tells the reader how fresh the figures are.
Length of 1,200 to 1,900 words suits most business topics. Shorter is fine if the question is narrow. Do not pad to hit a count.
How do you write the headings?
For articles, headings can be the reader's real question: "How long does a build take?" This is a retrieval aid. Assistants and search engines match question wording to question headings.
Rules:
- one question per heading, in the words a buyer would use;
- no clever or abstract headings; "Costs" beats "Where the money goes";
- the first sentence under the heading answers it;
- no heading that asks and then dodges.
Interface headings in a product stay as plain noun phrases. The question form is for articles.
How do you handle facts and numbers?
This is where most articles lose trust.
- Every percentage, price, deadline or statutory rate gets an inline link to the authority: IRAS, ACRA, MOM, CPF Board, MAS, or the vendor's own pricing page.
- Check the figure at the source on the day you write. Do not copy it from another blog.
- If you cannot confirm a number, leave it out. A vague true sentence beats a precise false one.
- Put the check date at the end.
- Name your own numbers precisely and say where they come from. "About S$0.05 per report, measured on our live systems" is usable. "Very cheap" is not.
Assistants tend to prefer claims they can attribute. Readers do too.
How do you keep the voice usable?
- Short sentences, one idea each.
- Active voice. "The agent logs every action", not "Every action is logged by the agent."
- Define a term once, at first use.
- No hype words such as superlatives or stock adjectives. They carry no information and cannot be checked.
- No claims about competitors you cannot support. Compare facts: features, published prices, limits.
- Use lists for steps and requirements, tables for comparisons, paragraphs for reasoning.
Read the article aloud. Where you stumble, cut.
What about schema markup and special files?
Structured data helps when it matches the visible page, and a clear title and meta description help in search results. Google's guidance, cited above, says no special schema is needed for AI features. A file such as llms.txt is a separate question; see llms.txt: what it is and whether to publish one. Neither replaces a clear first paragraph.
How do you check the article worked?
Use the method in how to check if ChatGPT, Perplexity and Gemini cite your site. Add the article's question to your fixed set of 20. Test before publishing to get a baseline, then at four weeks and eight weeks. Watch three things: whether your brand is named, whether your URL is cited, and whether what is said about you is correct.
If the article is indexed and still not cited, compare it with the page that is cited. The difference is usually in the opening, the specificity or the sources.
What does a rewrite look like?
Take an illustrative opening that fails the answer-first test, and fix it in three passes.
Before: "Many businesses are asking whether to automate their payroll. There are several factors to think about, and the right choice depends on your situation."
Pass 1, state the answer: "Automate payroll once you run more than one pay cycle with changing inputs, such as variable hours or allowances. Below that, a checked spreadsheet is enough."
Pass 2, add the condition that decides it: "The deciding test is whether one person can re-check every payslip in an afternoon."
Pass 3, add the source for any statutory figure, such as a CPF contribution rate, with a link to the CPF Board page, and leave the figure out if you cannot confirm it.
Final paragraph: three sentences, one claim, one condition, one test. Nothing in it depends on the rest of the article.
A quick audit for any existing page:
- Read only the first paragraph. Does it answer the title?
- Read only the H2s. Do they form a plausible outline of the answer?
- Read only the table. Does it stand alone?
If all three read well on their own, the page is easy to quote accurately.
Where KYN fits
KYN builds an article engine that follows this structure. It drafts from a brief, adds the sourced figures with links, and holds the draft for a named person to approve before anything is published. It costs about S$0.06 per article on our own systems. Everything runs in your own cloud account, and every action is logged with the actor. See marketing systems.
The first 14 days are a review of your stack and interviews with your team. A full build ships in under 90 days. For the wider context, read what is generative engine optimisation.
Frequently asked questions
How long should the answer paragraph be?
Two to four sentences. Long enough to carry the claim and one condition, short enough to quote in full.
Should I put the answer in a summary box?
You can, but the first paragraph itself should already be the answer. A box that repeats a different, vaguer opening adds nothing. Put the answer in the text where every reader and every crawler sees it first.
Can AI write the article for me?
It can draft one. A person must check every fact against its source, remove claims the sources do not support, and approve it. Unreviewed drafts are the common source of wrong figures.
Checked 30 October 2026.
Curious whether Marketing agents fits your business? Talk to KYN on WhatsApp — no forms, just a conversation.
