AEO Checker

Question Headings Are Cheap to Fix. Here's What the Evidence Says They Do, and What Google Says They Don't Need.

Cal Evans 2026-10-02 7 min read

Two people telling you opposite things

Picture a plumber, call him Dave, who gets two pieces of advice about his website in the same week. A vendor says every page needs answer passages of an exact length, formatted just so, or the AI engines will skip him. Then he reads that Google has published a guide saying you don't need to chop your content into tiny pieces at all.

Both can't be right. Dave closes the laptop and goes to fix a water heater, which is probably the correct call.

Here's how I sort it out.

What Google actually said

Google's generative-AI search guide, last updated July 10, 2026, is blunt. "There's no requirement to break your content into tiny pieces for AI to better understand it." And: "You don't need to write in a specific way just for generative AI search."

It also says structured data isn't required and there's no special schema.org markup to add. On llms.txt-style files: "You don't need to create new machine readable files, AI text files, markup, or Markdown to appear in Google Search (including its generative AI capabilities), as Google Search itself doesn't use them."

Search Engine Journal's Matt G. Southern covered the May version of the guide and quoted the line that sums up the attitude: "From Google Search's perspective, optimizing for generative AI search is optimizing for the search experience, and thus still SEO."

The same guide also says "People generally appreciate it when web pages are organized by paragraphs and sections, along with headings that provide a clear structure to navigate content." Google isn't against headings. It's against the idea that you need a special trick.

And all of it covers Google Search. It says nothing about ChatGPT, Claude, Perplexity, or Copilot.

What the studies found, and what they can't prove

Two pieces of research are worth knowing about. Neither one is a rulebook.

Kevin Indig analyzed ChatGPT citations: 1.2 million responses and 18,012 verified citations. The original is paywalled, so these figures come from secondary coverage, via Search Engine Land and a write-up by ALM Corp. 44.2% of citations came from the first 30% of a page, 31.1% from the middle 30 to 70%, and 24.7% from the final 30%. Within paragraphs, 53% came from the middle, 24.5% from the first sentence, and 22.5% from the last. The same reports say 78.4% of citations that contained a question came from headings.

That last figure describes where question text sat when a citation happened. It doesn't show that adding a question heading causes a citation. And it's ChatGPT only.

MaxAEO, a vendor, published its own study on July 17, 2026: 3,200 cited passages across eight engines, 600 informational queries, about 60 days in early 2026. Its finding was that 64% of quoted passages were the first or second sentence of their section or paragraph, with a median length of 28 words. The authors call their own work "observational, not causal" and "directional, not laws." The query set is B2B SaaS, marketing, and tech, not a homeowner searching for a plumber.

Put the two side by side and the numbers look like they disagree: 24.5% versus 64%. They don't. One counts the first sentence of a paragraph in ChatGPT. The other counts the first or second sentence of a section or paragraph across eight engines. Different units, different samples, not comparable.

What both do point at, loosely, is the same direction: the answer tends to sit early. That's consistent with what Google says about readers appreciating clear structure. It's also about as far as the evidence goes.

Numbers that don't trace

You'll also run into "passages of 40 to 75 words get cited 3.1x more often" and a "2.5x extraction lift" for question-format headings. I couldn't find a primary source for either. One question-heading article I read cited Semrush, zero-click, and Bain statistics with no links to any of them.

Untraceable isn't the same as false. It does mean you shouldn't rewrite your site around it, and you shouldn't repeat it to a client.

What the Content Structure module checks

Our Content Structure module is 20 points, and it rewards a structure Google says isn't required for Google Search. The case for it rests on three things: human readers, other answer engines, and observational studies like the two above.

The five checks:

  • Question headings (3 points). H2 or H3 headings containing a question mark. Three or more passes, one or two is a warning.
  • Questions answered directly (6 points). The first thing under each question heading has to be a paragraph of at least 20 characters, with a first sentence of 50 words or fewer, that doesn't open with filler like "In this article" or "When it comes to." Pass at 80% answered, warning at 40%.
  • Extractable answer format (3 points). The answer paragraph is immediately followed by a list or table. Pass at 50% of answered questions.
  • Word count (6 points). On a multi-page audit, an average above 400 words per page passes and 200 to 400 is a warning, and if more than 40% of pages are under 300 words it drops a level. On a single page, more than 300 words passes and 150 to 300 is a warning.
  • Visible FAQ section (2 points).

There's no word-count window for the answer, and no formula. A 50-word ceiling on the first sentence is a readability check, not a magic number. If you want the schema side of this, I covered it in Google Killed FAQ Rich Results, and Your FAQ Schema Still Has a Job.

When I ran our own site through the audit in I Took My Own Medicine, aeocheck.net scored 17 out of 20 here. Six question headings, all answered directly, three followed by a list or table, a visible FAQ. The three points I lost were all word count: pages averaged 311 words and two of six were under 300.

A 10-minute check for your top service page

Pick the page that earns you the most money. Drain cleaning, AC repair, whatever. Then:

  1. Write down three questions customers ask you on the phone. Not keyword research. The actual questions. "How much does it cost to replace a water heater?" "Do you do same-day service?"
  2. Turn each into an H2 or H3 heading with a question mark.
  3. Directly beneath each heading, write a plain answer in one or two sentences. Lead with the answer. Something like: "Most water heater replacements take two to four hours, and we can usually do it the same day you call." If you have a list of what's included or a table of price ranges, put it after that paragraph, not instead of it. Don't open with the list.
  4. Scroll through the page and check whether your most important answers are buried below roughly the first 30% of it. In the Indig data, the first 30% of the page drew more ChatGPT citations than any other section (44.2%), though not a majority. Our module doesn't measure this, so it's on you.

That's it. It costs you nothing but a few minutes and a little honesty about what people really ask. Even if every AI study turns out to be noise, a customer scanning your page on a phone, stressed, with water on the floor, will find the answer faster. Google says people appreciate that. Dave's customers would too, and he'd never have to read another vendor formula.

Find out where you stand

The free Grade Check at aeocheck.net scans one page and gives you a 0 to 100 score and a letter grade off the same engine behind the paid reports. It takes about a minute and costs nothing. It won't walk through the five checks above. For that, run your own ten-minute version first, then see what the number says.