Reading time: 13 minutes
To write a blog post that Google and AI cite, put the answer to the main question in the first paragraph. Then build the blog post structure: sections with clear headings, data with sources, tables for comparisons, and a visible author and date. This guide shows each step, as well as what the studies say about it.
In short
- In a Semrush study, an answer or summary at the start of the text had the strongest association with AI citation.
- In the CXL sample, 55% of the passages cited by Google AI Overviews were in the first 30% of the page.
- Google says you don’t need to break text into small chunks for AI, and that there is no ideal page length.
- Show the author, the source of each number and the update date, but change the date only when the content really changes.
What is the ideal blog post structure?
The ideal blog post structure starts with a descriptive title and the answer in the first paragraph. Then come the summary box, the sections with clear headings, the data with sources and the tables. Finally, the author and date stay in view. Each block has support in Google’s documentation or in citation studies.
| Block | What to do |
|---|---|
| Title | Say what the text delivers |
| First paragraph | Answer the main question |
| Summary box | Gather 3 to 5 key facts |
| Sections | One clear heading per topic |
| Data | Number with source and link |
| Tables and lists | Comparisons and steps |
| Author and date | Byline and “updated on” |
For Microsoft, which owns Bing, the title should sum up in natural language what the page delivers. The H1 should also reflect it. Google asks a similar question in its content self-assessment: does the title give a descriptive, helpful summary of the text?
Before you write the blog post, however, pick the right question. The guide on how to do content marketing shows how to plan your topics.
Where should the direct answer go?
In the first paragraph, that is, before any context. So the first sentence answers the main search, while the next ones say why it matters and who the text is for. In citation studies, the first third of the page holds more cited passages than the middle or the end. Also, a summary at the top had the strongest association with citation.
Semrush compared 304,805 pages cited by ChatGPT, Google AI Mode and Perplexity with 921,614 from Google’s top 20 for related searches. Among 13 text traits, the one most linked to citation came at the start. It was a summary, a key takeaway or a direct answer before any explanation. The gap between the groups was 32.83%. Semrush sells SEO and AI tools, and the study shows association, not cause.
CXL, which sells marketing courses, mapped 100 citations in Google AI Overviews. Of that total, 55% of the cited passages were in the first 30% of the page. The sample is small. However, CXL points to a similar pattern in Kevin Indig’s analysis of 18,012 ChatGPT citations. There, 44.2% came from the first 30% of the text.
Answering at the start of each section also helps people who skim. In the Semrush study, however, that per-section answer barely separated cited pages from non-cited ones (-1.1%). So when you write a blog post, treat the first paragraph as a complete answer.
How long should paragraphs and sections be for AI to cite a passage?
There is no magic length. After all, Google says you don’t need to split content into small pieces for AI, and that there is no ideal page length. Therefore, one section per topic works, with a clear heading and short paragraphs. On the other hand, very small blocks of a few lines showed no advantage in the studies.
This split into short, standalone blocks got the name content chunking. According to Google’s guide to generative AI features, its systems understand several topics on the same page and show the relevant part. At the same time, the guide asks for paragraphs, sections and headings that give a clear structure to navigate.
In the Semrush study, a clear heading for each topic had a 22.91% association with citation. By contrast, passage independence (3.64%) and one idea per section (-1.1%) made almost no difference.
SE Ranking, which sells ChatGPT tracking, analyzed 216,524 pages across 20 niches. Pages with average sections of 120 to 180 words between headings had 4.6 ChatGPT citations on average. With sections under 50 words, they had 2.7. However, pages with sections over 180 words had 5.7, which the company links to more complete coverage.
Thus, in your blog post structure, we recommend sections of 120 to 250 words and paragraphs of up to four lines.
Do question headings help a blog post get cited?
They help the reader and may help AI, but the studies disagree. That’s because a question heading repeats the searcher’s doubt and marks where the answer starts. So, when you write the blog post, use a question when the section answers a real doubt. Use a statement when it gives a step.
Microsoft, for example, prefers a heading like “What Makes This Dishwasher Quieter Than Most Models?” over a vague heading such as “Learn More”. Headings, it says, mark where one idea ends and another begins.
In the Semrush study, the question-and-answer format had a 25.45% association with AI citation. In SE Ranking’s data, however, pages with a question-style title or H1 had fewer ChatGPT citations, 3.4 versus 4.3. Still, the company says the effect depends on context. It also suggests using question-style subheadings naturally.
There is also query fan-out: Google splits the question into related searches. As a result, a section that answers one of those sub-questions can become a source, even if the page doesn’t rank for the original search. The mechanism is in our guide to Google AI Mode and AI Overviews.
Do tables, lists and a summary box make a difference?
They do, as long as they organize content that is already good. A table is for comparing, a numbered list is for steps and the summary box is for the key facts. In the studies, cited pages use these formats more often, but that’s association, not cause.
In fact, in the Semrush study, lists, tables, bullets and question-and-answer pairs had a 21.60% association with citation. Trakkr, which sells AI tracking, analyzed 1,465 pages cited by ChatGPT, Perplexity and Gemini. Among the 10% most cited, 40% had a table, versus 28.2% of the least cited half. However, the study doesn’t compare them with pages that AI didn’t cite.
Microsoft notes that lists, numbered steps and comparison tables break complex details into reusable segments. But it also asks for restraint: bullets work for steps, comparisons and highlights, not for every line. Cecilia Meis, a senior editor at Semrush, sums up the logic:
“Clarity and structure are not SEO shortcuts. They simply make information easier for both people and AI systems to interpret.”
Cecilia Meis, Semrush (2026)
FAQs at the end of the text also work as blocks that make sense on their own. In the CXL sample, some of the citations from the end of the page came from FAQ blocks. To build one, see how to create an FAQ that Google and AI use.
Author, sources and date: what should a blog post show?
Show who wrote it, where each number comes from, and when the text came out and changed. After all, Google asks for a byline where readers expect one and a clear source. Besides building reader trust, authorship and sourced data show up more on the pages that AI cites.
In the Semrush study, for example, showing authorship and credentials had a 30.64% association with AI citation.
Anyone who sets out to write a blog post should name the source of each data point and link to the original. In SE Ranking’s data, pages with expert quotes had 4.1 ChatGPT citations, versus 2.4 without them. Those with 19 or more statistics had 5.4, versus 2.8 for those with little data. But the company itself says these factors weigh less than others. The effect of citing sources and adding statistics was measured in the study that coined the term GEO. It is in our guide on what GEO is.
For the date, Google’s page on byline dates recommends a prominent, labeled date, such as “Published” or “Last updated”. The same date should also appear in the article’s structured data. In short, author, sources and date close out the blog post structure.
Why does the post need something of its own?
Because AI combines several sources, and a text that only repeats what already exists adds nothing to the answer. For this reason, Google asks for a unique perspective and first-hand experience. So, before you write a blog post, list what only your team knows.
In its generative AI guide, Google asks you not to recycle what others have already said. Nor should you publish what an AI model could easily produce. Likewise, one of Google’s self-assessment questions is whether the content provides original information, research or analysis.
Two formats do this with method. Original research turns company data or a survey into a number other sites can cite. Meanwhile, a case study shows a client’s result with context, timeframe and numbers.
How to write a blog post: step by step
To write a blog post with this structure, follow the order below. It works for a new text and for revising an old one. The length ranges are our own recommendation; the rest follows Google and the studies above.
- Pick a real question: for example, from Search Console, from “People also ask” or from sales and support questions.
- Write the title: say what the text delivers, in natural language and without hype.
- Answer in the first paragraph: we recommend 40 to 60 words, with the answer in the first sentence.
- Build the summary box: gather 3 to 5 key facts, each in a full sentence.
- Split the text by sub-questions: one H2 per topic, phrased as a question whenever the section answers a doubt.
- Open each section with the answer: two or three direct sentences, then the detail.
- Put a source on every number: name the source and link to the original.
- Use a table to compare: keep the numbered list for steps.
- Bring something of your own: for example, a data point, a test, a case or the view of the person who did the work.
- Show the author and date: finally, check every fact before you publish.
How do you revise an old blog post for this format?
Start with posts that already get visits or impressions but don’t show up in AI answers. In each one, rewrite the first paragraph with the answer, swap vague headings for real questions, and update numbers and sources. Then add a table where there is a comparison, and only then change the date.
Semrush itself suggests starting with pages that rank well on Google but perform poorly in AI search. For those pages, the process is the same as when you write a new blog post. The difference is that the blog post structure changes without losing the URL and its history.
The date, however, needs care. Among the signs of content made for search engines, Google lists changing a page’s date without substantially changing the content. Meanwhile, the habit of updating content is associated with more citations. In SE Ranking’s data, pages updated in the last three months had 6 ChatGPT citations on average, versus 3.6 for outdated ones.
To find out whether AI started citing the post, run the same questions several times, before and after the revision. The method is in our guide on how to measure brand mentions and citations in AI. But if AI keeps citing another site, the cause may lie outside the text. See why AI doesn’t cite your brand, from access failures to third-party sources.
What we found when we analyzed 10 pages on how to write a blog post
We ran searches in Portuguese on how to write a blog post and analyzed 10 Portuguese-language pages that came up. Only one treats AI answers, such as ChatGPT and AI Overviews, as a place where the post can appear. In fact, it is the page of a company that sells an AI blog writer.
Likewise, none recommends question-style headings. Only three ask for the answer or a summary at the start of the text or section. Four recommend linking sources for credibility, two mention tables and only one suggests showing the author profile and the date.
Three set word-count targets, which range from 600 to over 2,000 words, although Google says it has no preferred word count. Finally, the last update of two of them was in 2024 or earlier. In short, the pages teach blog post structure for the human reader and say almost nothing about AI.
Frequently asked questions
There is no right number. In fact, Google says it has no preferred word count. The studies disagree: at Ahrefs, more than half of AI Overview citations went to pages under 1,000 words. On the other hand, in SE Ranking’s data, texts over 2,900 words had more ChatGPT citations than those under 800.
Yes, and the path is the same as good SEO. According to Google, you don’t need to write in a specific way for AI search. The studies associate the blog post structure in this guide with citation: answer first, sections with clear headings, sources, tables and visible authorship. But none of this guarantees a citation.
No, it doesn’t. According to Google, structured data isn’t required for generative AI search, and there is no special markup for it. Also, in SE Ranking’s data, pages with FAQ schema had even fewer ChatGPT citations, 3.6 versus 4.2. Even so, keep the article schema your SEO plugin already generates, because it helps with rich results.
Want to know how your brand shows up in AI answers today?
NotHuman is opening spots for agencies and brands to track how they appear in AI answers. Join the waitlist and we’ll let you know when a spot opens up for you.
Main sources
- Google Search Central, guide to optimizing for generative AI features in Google Search (2026).
- Google Search Central, creating helpful, reliable, people-first content (2026).
- Google Search Central, influencing your byline dates in Google Search (2025).
- Semrush, study on the text traits linked to AI citation (2026).
- CXL, analysis of 100 Google AI Overview citations (2026).
- SE Ranking, study on ChatGPT citation factors (2025).
- Microsoft Advertising, how to prepare content for AI search answers, by Krishna Madhavan of Bing (2025).
- Trakkr, anatomy of pages cited by AI (2026).
