Anchor answer: the passage an AI copies from your page
An anchor answer is the short paragraph, up to roughly 45 words, placed directly under a question-form heading and written to answer that question without depending on the rest of the page. It is the unit the retrieval layer can cut out and return inside a generated answer.
What is an anchor answer?
An anchor answer is the short paragraph placed directly under a question-form heading, answering that question in full before any elaboration. It is the passage prepared to be retrieved on its own and cited by an AI assistant.
The anchor answer comes from an observation about how retrieval works. An AI assistant does not return pages, it returns passages, picked through query fan-out and recombined into a single answer. What competes for the citation is the passage, with its section heading travelling alongside it.
Hence the pair that structures every ReBo AI-friendly page: an H2 in question form mirroring a target prompt, with the anchor answer that satisfies it right below. The rest of the section is there for whoever kept reading.
How long should an anchor answer be?
The ReBo standard is up to roughly 45 words, in declarative sentences. The cap exists because the passage has to fit whole into the space a generative engine gives a source. Below about 20 words the answer tends to be too vague to be useful.
| Parameter | ReBo standard | Why |
|---|---|---|
| Length | up to ~45 words | fits whole into the space an engine gives a source |
| Position | first paragraph after the H2 | the model reads the question and answer as one unit |
| Sentences | declarative, averaging 25 words or fewer | long sentences dilute the claim and hinder extraction |
| Opening | no 'this', 'moreover', 'as we saw' | the passage is read out of the page's order |
| Content | names, figures and dates instead of generalities | factual density is a selection factor |
Source: AI-friendly page technical standard, ReBo AI Visibility, September 2026.
The word cap comes with an honest counterweight. Page length barely correlates with citation, and more than half of the pages ChatGPT cites run under a thousand words, according to the Ahrefs measurement published in August 2026. What the anchor answer optimizes is the density of the passage, without inflating the text around it.
Why is that the passage a model copies?
Because it satisfies the sub-search by itself. During retrieval, the system looks for a passage that answers the question without depending on the rest of the page. A paragraph that opens with context and reaches its conclusion at the end forces the model to assemble the answer itself.
Compare two openings for the same section. The first begins with "In recent years, much has been "written about the role of structured data", and the useful claim arrives in the fourth sentence. The second begins with "Structured data removes ambiguity for machines, and on its own it does not earn a citation". The second can be cut out; the first has to be read whole to yield anything.
Verifiability weighs as much as form. A passage carrying a figure, a date and a declared source gives a model what it needs to stand behind the citation, while a passage of adjectives and superlatives does not. That is why anchor answers sit next to sourced tables and an authorship box in the ReBo method.
Related entries
An anchor answer responds to a target prompt and is cut out during retrieval, which query fan-out multiplies. Its cumulative effect shows up in the citation rate. Before any of that, an AI crawler has to be able to fetch the page.
Frequently asked questions
Is an anchor answer the same as a featured snippet?
They are relatives with different destinations. A featured snippet is a slot on Google's results page, extracted by classic search. An anchor answer is written for the retrieval layer that feeds generated answers, where what counts is a passage that is self-contained and verifiable.
Does every section need an anchor answer?
Every section that opens with a question does. Narrative sections, such as the timeline of a case or the description of a process, work under the simpler BLUF rule: the main information in the first sentence, the elaboration afterwards.
Can the anchor answer repeat what the FAQ says?
It can, and usually does, on one condition: the FAQPage text in the JSON-LD has to match what is visible on the page word for word. Broken parity is a contradictory signal, and the cost lands on trust in the whole source.
How do you tell whether an anchor answer is working?
By running the page's target prompt on the engines and logging what gets cited. When the anchor answer works, the cited passage shows up almost verbatim in the generated answer. That is citation measurement, rather than position in a list of links.