Answer Block Scorer
Paste a draft and see whether it opens with a passage an answer engine could lift — scored on the same signals the Citely audit uses.
Analysed in your browser — an unpublished draft never leaves this machine.
Engines quote passages, not pages
This is the distinction most writing advice misses, and everything else here follows from it. When an assistant answers a question, it does not summarise your article — it lifts a passage out of it and attributes that. A page can contain exactly the right answer and never be used, because the answer only makes sense after three paragraphs of context that came with it.
The fix is structural rather than stylistic: put a complete answer at the top, short enough to stand alone.
What counts as an answer block
Roughly 10 to 120 words, at the very start, that answer the question implied by the title without depending on anything around them.
- The floor exists because a one-line kicker, byline or date is not an answer, and without a minimum every page with a short opening line would score perfectly.
- The ceiling exists because a passage that needs editing before it can be quoted generally is not quoted.
A useful test: could you paste your first paragraph into a reply to someone who asked the question, without changing a word and without the rest of the page? If not, it is an introduction rather than an answer.
The other signals, and why they are weighted as they are
The scoring here follows the published research rather than intuition. Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024), tested a range of interventions and found citing sources, adding quotations and adding statistics produced the largest gains among those tested. So:
- External citations — distinct external domains, counted once each. Ten links to one partner is not broad sourcing, and a page citing itself corroborates nothing.
- Quotable figures — percentages, amounts, magnitudes, multipliers. A specific number is far more repeatable than a qualitative claim, because an engine can attribute it precisely.
- Question-shaped headings— people prompt in questions, and a heading reading “How do I verify this?” matches a prompt more closely than “Verification”.
- Substance — under about 300 words, a page is rarely chosen when a fuller source is available on the same question.
- A modified date — a weak signal alone, but a page with no date is hard to trust for anything time-sensitive.
What the score is, and is not
It is a coarse ranking device for comparing your own drafts against each other. The weights are a judgement call informed by the research, not a published formula, and there is no threshold above which anything is guaranteed.
No score predicts a citation. That depends on the prompt, the other documents retrieved alongside yours, and ranking behaviour nobody outside those companies can observe — and it varies between runs of the same prompt. What this measures is whether you have given an engine something usable, which is a precondition rather than a cause.
The same analysis the plugin runs
This scorer and the Citely plugin’s extractability audit share one implementation and one set of thresholds, held in step by a shared fixture set that both test suites read. A draft that scores well here scores the same once it is a published post — which is the point, since finding out at draft stage is when the fix costs a sentence rather than a rewrite.
When structure is not the problem
If a page is well-structured, well-sourced and still never appears, the constraint is usually reachability rather than writing. A blocked retrieval crawler removes you from an assistant’s answers entirely regardless of how quotable the page is — check that with the AI crawler checker, and read training vs. retrieval crawlers for why it is so easy to do by accident.
Frequently asked questions
- What is an answer block?
- A short, self-contained passage — usually the opening paragraph — that completely answers the question implied by the title without needing the surrounding context. Answer engines assemble responses from passages they can lift whole, so a page whose answer only makes sense after three paragraphs of preamble may never be quoted even when it is the best source available.
- How long should an answer block be?
- Roughly 10 to 120 words. The floor exists because a one-line kicker or byline is not an answer; the ceiling because a passage long enough to need editing before it can be quoted usually is not quoted. Within that range, shorter and more specific beats longer and more careful.
- How is this different from the AI Visibility Checker?
- The visibility checker fetches a URL, so it only works on pages that already exist. This scores text you are still writing, which is when the findings are cheapest to act on. Both run the same analysis with the same thresholds, so a draft that scores well here will score the same once published.
- Does a high score mean I will get cited?
- No. The score ranks your own drafts against each other and nothing more. Whether an engine cites a page depends on the prompt, the competing sources and ranking behaviour nobody outside those companies can observe. What the score measures is whether you have given an engine something it could use — a precondition, not a cause.
- Why does it want me to cite external sources?
- Because that is what the published research supports. Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024), tested a range of content interventions and found citing sources, adding quotations and adding statistics produced the largest visibility gains among those tested. Distinct external domains are counted once each, so a page linking the same partner ten times does not read as broadly sourced.
- Is my draft uploaded anywhere?
- No. Parsing and scoring both happen in your browser. That is deliberate — an unpublished draft is exactly the kind of thing that should not be posted to somebody else's server to be analysed.
Structure is only half of it — a perfectly quotable page still needs a crawler to be allowed to fetch it.