Fixing Content Structure for AI Extraction
AI assistants rarely quote a whole page. They pick out a passage, often a paragraph or a few list items, and build their answer from it. If your answer is spread across several paragraphs, hidden behind a vague heading, or only makes sense after reading what came before, the assistant either misses it or quotes something less useful.
This guide shows how to find those problems on your own pages and restructure them, section by section, so each important answer can be lifted cleanly and still be correct.
How assistants take passages from a page
When an assistant searches the web to answer a question, it does not read your page the way a visitor does, top to bottom with the design around it. Its systems split the text into pieces, usually along headings and paragraphs, and compare each piece with the question. The best-matching pieces are handed to the model, which writes the answer and, in many assistants, adds a link to the page they came from.
No assistant publishes exactly how it divides pages, and the details change. But the practical lessons are consistent across all of them:
- Headings mark boundaries. A clear heading tells the system where one topic ends and the next begins, and what each piece is about.
- The first sentences carry the most weight. A section that opens with the answer matches the question closely. One that opens with background may be passed over.
- Pieces are read without their neighbours. A paragraph that depends on the one before, with phrases such as "this" or "as mentioned", may be understood wrongly or not used.
- Structure in HTML survives; structure in design does not. A real list or table keeps its meaning as text. Bold text styled to look like a heading, or a comparison drawn as an image, does not.
None of this is at odds with writing for people. Readers skim headings, want the answer first and dislike hunting through a page. A page structured for extraction is almost always easier to read.
A section before and after
Here is a typical section from a fictional dog groomer's services page, and the same content restructured. The facts are identical.
Before:
<h2>Our Approach</h2>
<p>At Example Grooming we believe every dog deserves the best.
Our team has years of experience and we love what we do.</p>
<p>It depends on the breed and coat, but for most dogs we
recommend coming in regularly. As mentioned above, some breeds
need more frequent visits, usually every four to six weeks,
whereas others can go eight to twelve.</p>
After:
<h2>How often should a dog be groomed?</h2>
<p>Most dogs need a full groom every six to twelve weeks. Breeds
with continuously growing coats, such as poodles, bichons and
shih tzus, usually need one every four to six weeks to prevent
matting. Short-coated breeds can often go twelve weeks or more
between professional grooms.</p>
<p>At Example Grooming in Derby we set a schedule for each dog at
its first visit, based on coat type and how it is cared for at
home.</p>What changed:
- The heading now says what the section answers, in the words an owner would use.
- The first sentence is a complete answer that would make sense quoted alone.
- The vague "regularly" and "some breeds" became named breeds and specific intervals.
- "As mentioned above" is gone; nothing depends on an earlier paragraph.
- The business and its location are named in a separate sentence, so a quoted answer can still be attributed.
- The paragraph of general claims about passion and experience was removed from this section. If it matters, it belongs on the About page, with evidence.
Finding structure problems on your pages
Start with the pages that should answer your customers' most important questions, not the whole site. For each one, run these checks.
1. Read the heading outline
Open the browser developer tools and inspect the headings, or use an accessibility inspector or outline
extension that lists h1 to h6 in order. Read the list on its own. Does it tell
you what the page covers and where each answer is? Headings such as "Overview", "More info", "Why us"
or a lone brand slogan say nothing about content.
2. Look for fake headings and fake structure
Page builders often let editors style a paragraph to look like a heading, or use a heading tag for
decoration, such as an h2 around a phone number. Search the page source for your visible
headings and check they are real heading elements at a sensible level. Check that lists are ul
or ol, not paragraphs with dashes, and that tables are table elements, not
images or grids of divs.
3. Find where each answer starts
For each question the page should answer, find the first sentence that actually answers it. If it is more than two or three sentences below its heading, or under a heading that does not name the question, the section needs restructuring.
4. Read key paragraphs alone
Copy an important paragraph into a blank document and read it cold. Does it make sense? Does it name what it is talking about? Does it contain words like "this", "they", "the above" or "as we said" whose meaning is on another part of the page?
5. Check what the assistants quote now
The AI Visibility Tracker records whether your brand is cited, mentioned, recommended or missing for your customers' questions. Where you are cited, open the answer and see which passage was used. Where a competitor is cited instead, compare the structure of their section with yours.
The fix, step by step
Step 1
Map each page to the questions it should answer
For each important page, list the three to eight questions it ought to answer, using the words customers use in enquiries, reviews and search queries. Remove questions another page on your site already answers better, and link to that page instead. If one page is trying to answer twenty unrelated questions, split it. The list becomes the plan for the page: each question will get its own section with its own heading and answer.
Step 2
Rewrite headings so they name the question or topic
Replace vague headings with ones that say exactly what the section answers.
A question works well when the section answers one question, for example How long does a boiler service take?.
A plain topic phrase suits reference content, such as Boiler service costs in 2026. Keep
one h1 for the page, use h2 for main sections and h3 inside them,
and never skip levels just to get a smaller font size.
Step 3
Put the answer in the first sentences of each section
Under each heading, write one to three sentences that answer the question fully before adding any background, caveats or selling. Include the specific figure, time, condition or step that makes it a real answer. Then follow with the detail a careful reader needs: exceptions, examples and how your business handles it. If the honest answer is that it depends, say what it depends on and give the typical range.
Step 4
Make every key passage stand on its own
Go through each answer paragraph and replace words that point elsewhere, such as this,
it, the above or as mentioned, with the thing they refer to. Name
the product, service or place in the paragraph. Where attribution matters, name your business once in
the section. Keep each paragraph to one idea, so a quoted paragraph cannot mix an answer to one question
with half of the answer to another.
Step 5
Turn comparisons and specs into real lists and tables
Find information presented as images, sliders, tabs built from scripts, or text lines separated by dashes. Rebuild steps as an ordered list, options as an unordered list, and comparisons or specifications as an HTML table with a header row. Give each table a short sentence before it saying what it compares. Keep units, currency and dates inside the cells, so a single row still makes sense when lifted out.
Step 6
Repair the heading outline in the template
Many heading problems come from the theme, not the content: a logo wrapped in h1
on every page, sidebar titles marked as h2, or a page builder that outputs styled div
elements instead of headings. Fix these in the template or builder settings so every page benefits. Afterwards,
list the headings on a few pages again and check the outline now reflects only the main content, in a
sensible order.
Step 7
Test passages out of context, then re-measure
Copy each rewritten answer into a blank document and read it without the page around it. Ask whether a stranger would understand it, whether it is correct on its own, and whether it says whose answer it is. Publish, wait until the page has been recrawled, then re-run your customer questions. Compare which passages are quoted and how often you are cited against your first measurement.
Structures that suit different kinds of question
Different questions need different shapes. Matching the shape to the question gives assistants a passage that fits their answer without rewording.
| Question type | Example | Structure that works |
|---|---|---|
| Definition | What is a party wall agreement? | One sentence defining it, a second saying when it applies, then detail. |
| How-to | How do I bleed a radiator? | A one-line summary, then an ordered list of steps, each starting with a verb. |
| Cost or time | How much does a website audit cost? | The typical range first, then what moves it up or down, with a date for prices. |
| Comparison | Combi or system boiler? | A one-sentence verdict by situation, then a table of differences. |
| Yes or no | Do I need planning permission for a garden room? | The answer and its main condition in the first sentence, then the exceptions as a list. |
| Choice | Which grooming package is right for my dog? | A short list pairing each option with the situation it suits. |
A page can mix these. A service page might open with a definition, include a cost section and end with a comparison table. The point is that each section has one job and one shape.
Platform notes
WordPress
In the block editor, use the Heading block and choose the level from its toolbar rather than making
a paragraph larger. The List and Table blocks output real HTML lists and tables. Check the theme: some
place the site title in an h1 on every page, and some page builders such as Elementor let
a heading widget output any tag, which editors may set to div or p for styling.
Set those widgets to the correct heading level.
Shopify
Product descriptions are usually written in the rich text editor, which supports headings and lists but is often left as one long paragraph. Specifications work well as a table in the description, or as product metafields displayed by the theme as a list. Check that tabs and accordions in the theme contain their text in the page HTML rather than loading it when clicked.
Static sites and frameworks
Markdown-based sites generally produce clean headings, lists and tables. Watch for components that
render headings as styled div elements, content inside tabs that only exists after a click,
and comparison tables built as CSS grids. Server-rendered or statically generated HTML keeps all of it
available to crawlers.
Over-correcting: mistakes to avoid
- Turning every heading into a question. Questions suit sections that answer one question. Reference material, such as specifications or terms, reads better under plain topic headings.
- Chopping pages into tiny fragments. A page with forty one-line sections is hard to read and gives no room for the detail that makes an answer trustworthy. Aim for sections a reader can finish in under a minute, not a single sentence.
- Adding an FAQ block to everything. A list of loosely related questions tacked onto the bottom of every page repeats content and dilutes the page's focus. Use an FAQ only where the questions genuinely belong to the page.
- Stuffing question phrases. Repeating "best loft conversion Sheffield" in headings does not make a passage more quotable. Answering the question does.
- Removing everything that is not an answer. Examples, evidence and first-hand experience are what make a page worth citing. Put them after the answer, not instead of it.
- Answer summaries that contradict the page. A short answer at the top must agree with the detail below. Update both when anything changes.
For how assistants attribute what they quote, and how to reduce misquotation, see fixing wrong attribution. For the structured data that sits alongside good structure, see fixing schema for AI answers.
Keeping new content well structured
- A page template for answers. Give editors a template with a question-led heading, an answer box of two or three sentences, then detail. It is easier to follow a pattern than a rule.
- A publishing checklist. Before publishing: headings name the content, each section opens with its answer, no paragraph depends on another, lists and tables are real HTML.
- Lock heading levels in the builder. Where the page builder allows it, restrict heading widgets to real heading tags so styling choices cannot break the outline.
- Review top pages twice a year. Re-read the pages behind your most valuable questions, update facts and dates, and check the structure has survived edits.
- Measure after changes. Track the same customer questions before and after restructuring so you can see which patterns work for your market.
1. Should every heading be written as a question?
No. Question headings work well when a section answers one specific question that people actually ask, because the heading then matches the way the question is phrased. For reference content, such as specifications, terms, opening hours or a list of services, plain descriptive headings are clearer. What matters is that every heading tells a reader, and an assistant, exactly what the section contains. Vague headings such as Overview or More information are the ones to replace.
2. How long should an answer passage be?
Long enough to answer the question completely, which is usually one to three sentences for the core answer, followed by as much detail as the question needs. There is no fixed ideal length, and Google says there is no minimum word count for pages. The test is whether the opening sentences would still be a correct, useful answer if they were quoted on their own, without the rest of the section around them.
3. Will restructuring my content hurt my Google rankings?
It is unlikely to hurt, as long as you keep the useful content and the page still covers its topic well. Clear headings, answers near the top and real lists and tables are consistent with Google's own guidance on helpful, people-first content. Keep the same address for the page, avoid removing sections that bring in search traffic without replacing them, and check Search Console after the change to confirm the page is still performing.
4. Do tables and lists really help AI assistants?
They help because they keep their structure when the page is turned into text. An HTML table tells a system which value belongs to which row and column, and a list separates steps or options clearly. The same information drawn as an image, laid out in columns with CSS only, or written as a run-on paragraph loses that structure. Use a table when readers need to compare several options against the same attributes.
5. What does self-contained mean for a paragraph?
It means the paragraph makes sense if someone reads only that paragraph. It names the thing it is about instead of saying this or it, does not rely on an earlier sentence for key facts, and states any condition that changes the answer. Assistants often take a single passage from a page, so a paragraph that depends on the one before it can be misunderstood, quoted incorrectly, or skipped in favour of a clearer source.
6. Is it bad to hide content in tabs or accordions?
Not in itself. If the text is present in the HTML and only visually collapsed, crawlers and assistants can read it, and Google treats it as normal content. The problem comes when the text is loaded by a script only after someone clicks, because crawlers do not click. View the page source or the rendered HTML and search for a sentence from inside a closed tab. If it is not there, the content needs to be included in the page.
7. Should I add a summary box at the top of long pages?
A short summary or key points box can help both readers and assistants, especially on long guides, because it gives the main answers in one place near the top. Keep it short, make sure every point is covered in more detail below, and update it whenever the detail changes. A summary that contradicts the body is worse than none, because an assistant may quote the outdated summary instead of the correct detail.
The AI Visibility Tracker asks AI assistants the questions your customers ask and records whether your brand is cited, mentioned, recommended or missing, so you can see whether restructured pages are being used and track progress over time.
Open AEO Fixes →