The 200 characters under your H1

Resoneo compared 534 pages that ChatGPT had cited against the snippet OpenAI's index had stored for each one, and the study found that stored snippet is cut just past 200 characters.

I read it on 2026-09-25.

That cut is the whole page, as far as the index is concerned.

I nearly got this wrong before I started. The brief I wrote for myself credited the stored-snippet measurement to Ahrefs, and I spent an hour on ahrefs.com before I accepted that the Ahrefs post measures something else.

So the number belongs to Resoneo, and Ahrefs has a different number that points the same way.

But first the part that matters for your site. Whatever sits directly under your H1 is what an answer engine holds about you between crawls, and most pages spend it on a category label, a date and a hero image's alt text.

Step 1: Count what the index keeps

Resoneo's method was simple. Fetch each cited page, pull the snippet the index stored for it, and line the two up.

Of the 463 pages with an H1, 387 snippets contained it, which Resoneo puts at 83.6 percent.

Resoneo reports a median H1 of 51 characters and about 7 characters before it, leaving 146 characters of your own words. Count them on your homepage.

And the snippet did not change with the question. Pages seen under several queries returned an identical snippet, which is why Resoneo calls it frozen at crawl time.

Seven regional versions of one press release, asked the same thing, got seven different anchors. The page picks the snippet, and the question does not.

I did not expect that.

I had assumed the engine re-read the page for each question. Search Engine Journal's write-up from 2026-08-11 was the first thing that made me go and check.

Step 2: Read the Ahrefs number next to it

Ryan Law at Ahrefs published four writing frameworks for AI visibility on 2026-05-08. The one that matters here is bottom line up front.

The Ahrefs post reports that 44.2 percent of citations come from the first 30 percent of a page's content, citing internal research, and that citation winners were almost twice as likely to use definitive phrasing, 36.2 percent against 20.2 percent.

So two studies with different methods land in the same place. One measures what the index stores and the other measures where citations are drawn from, and both say the top of the page carries the weight.

That resolved my complication.

I had been treating the two findings as competitors, and they are two views of one behaviour.

Step 3: Write the block in this order

The pattern is four slots, and the order is fixed by how the budget runs out. Order matters more than polish.

Entity first, because if the cut lands before your name the snippet is about nobody.

Then what it does, in the plainest verb you have. Then one number you can source, then the next step a reader would take.

Four slots, in roughly 150 characters after the H1 by Resoneo's count. Characters, not words.

Write it as one or two flat sentences. Definitive verbs beat hedged ones, which is the Ahrefs finding again.

Step 4: Template one, a place people visit

[Name] is a [type of place] on [street or area], open [days and hours], known for [one thing], and you can [book or walk in] at [link].

Every bracket is a fact you already have.

The only one that hurts is "known for one thing", because most owners want to list six. Six is a list.

One is a claim, and the answer engine can only hold one clause about you, so decide which clause a stranger would repeat.

And keep the hours as digits. Words like "early" or "late" cost characters and carry nothing.

Step 5: Template two, a page that answers a question

[Question restated as a statement], [the direct answer in one sentence], [the number or date that supports it, with its source named], [what to do with it].

This is the template for a blog post, a help article or a pricing explainer.

The question goes first because the stored snippet is query independent, so the page has to declare its own question.

I did not know if it would work when we started writing posts this way.

The honest reading is that Resoneo did not test whether changing the block raises citation rates. The study measured what is stored, not what wins.

So the case rests on the two studies agreeing, and on the cost of the change being close to zero.

Step 6: Template three, a service or plan

[Name] does [service] for [who it is for], [scope in one clause], from [$ amount] a month, and the [next step] is at [link].

The price slot is the one most sites leave empty. We wrote about what that costs in hiding your prices keeps you out of AI answers.

If you will not publish a price, replace the slot with a sourced number about the work and say so plainly. Say so in the block.

Do not fill the number slot with a number you cannot defend. An empty slot is a smaller loss than a wrong one.

Step 7: Run the check on every page you have

Here is the check NovaTechRay runs first on any site. It needs no tool.

Open the page, find the H1, and copy the rendered text from the H1 forward until you reach Resoneo's 200-character cut, H1 included.

Read what you copied out loud.

Then ask three questions: does it name the entity, does it say what the entity does, and would a stranger know what to do next?

Two of three is a pass. Fewer is a rewrite.

Every failing page I have looked at fails the same way, with a kicker, a date and an image caption sitting where the answer should be.

Step 8: Strip what does not belong

Resoneo found the two real losses were the section kicker, present on 29 percent of pages at 18 characters, and the alt text of the first image, which the study says can eat fifty characters on its own. Fifty is a lot.

Skip the meta description.

That sounds reckless, so here is the defence. Resoneo reports the stored snippet is taken from the rendered body and that this index ignores the meta description, and Google's snippet documentation, read on 2026-09-25, says snippets are primarily created from page content and the description is used only when it describes the page better.

So write the description, but write it last, after the block under the H1 is right, because the block is what both systems read first.

Also out: a welcome line, a breadcrumb rendered as text, a date stamp above the title, and any sentence that starts with "In this article".

Step 9: What this does not fix

This block is a reading fix. It only decides what an engine holds about a page it has already found.

It does nothing for pages it has not found, and we covered how much of that is plain Google ranking in how much of an AI citation is just your Google ranking.

A perfect block on a page outside the top ten is a perfect label on a closed door.

But it is the cheapest change on the list, and NovaTechRay puts it first on every audit for that reason.

Resoneo also notes that one in seven cited pages in its set had no H1 at all. On those the snippet anchors wherever the template happens to start.

Mark up the H1. Then spend what is left on purpose.

The short version

• Copy the first 200 characters of rendered text from your H1 onward and read them out loud.
• Name the entity, say what it does, give one sourced number, name the next step.
• Move kickers, dates and the first image below the block.
• Write the block before the meta description, not instead of it.
• Publish the price when you can; fill the slot with a sourced number when you cannot.
• Fix ranking and crawl access first on pages that have neither.

This is what we do at novatechray.com.