Answer engine optimisation in order: a gated plan for one morning, not another explainer
A prompt that turns answer engine optimisation into a gated sequence, where each step has a condition that must be true before the next is worth doing.
- Works in
- ChatGPT, Claude, Gemini
- You need
- Your domain and one page you care about · The raw word count of that page from curl · Which AI crawlers your robots.txt allows · How many hours you actually have
- Written for
- how to do answer engine optimization
Scored by our own engine
This page, run through the audit we sell. Measured 4 August 2026.
This is not an explainer. The definitions are contested, nobody owns either label, and none of that helps somebody with a site and a free morning. Answer engine optimisation has an order, and each stage is close to worthless until the one above it is true. The prompt below turns that order into a plan with gates, cut to the hours you actually have.
The order, and why gates matter more than the list
Access, then response, then structure, then identity, then discovery, then what other sites say. Most published checklists contain roughly these items in roughly no order, which is how sites end up with beautifully rewritten pages that a crawler was never allowed to fetch.
A gate is a condition, not a task. Stage two does not begin until a plain fetch of your page returns real text. Stage three does not begin until stage two passes. That sounds obvious written down and it is not what happens in practice, because rewriting a paragraph is more appealing work than reading a robots.txt file, and the appealing work is at the wrong end of the sequence.
The gates also make the plan honest about what it does not know. If you have not measured your raw HTML word count, the prompt is required to mark that stage UNKNOWN and give you exactly one action: go and measure it. It is not allowed to infer a pass, which is the failure mode of every generic audit you have been sent.
What is documented, and what is a reasonable inference
Stage one is documented. OpenAI, Anthropic, Perplexity and Google all publish which crawlers they run and how robots.txt controls them, so whether a crawler may fetch you is a fact you can check rather than a theory.
Stage two is documented only in the negative. The three assistant operators publish nothing about whether their crawlers execute JavaScript, which is not the same as confirming they do not. Designing for a plain fetch is a decision made under that uncertainty rather than a fact.
Stages three and four are inference. A retrieval system chunks documents, so a passage that stands alone is a better candidate than one that does not. No operator has published that. It is worth doing anyway because the cost is an edit and the same edit helps a reader, which is the only kind of inferred advice worth taking.
Measure stage three with the AI content readiness check, and take the map rather than the sequence from the overview.
You are running a work session, not writing an explainer. I have one site, a
fixed number of hours and no interest in what the term answer engine
optimisation means. Produce an ordered plan with gates.
The order is fixed and you must not reorder it, because each stage is
worthless until the one above it is true:
1. ACCESS. Can a crawler fetch the page at all.
2. RESPONSE. Is there readable text in the raw HTML before JavaScript runs.
3. STRUCTURE. Is the page broken into sections a retrieval system can chunk,
with a standalone answer opening each one.
4. IDENTITY. Is it clear who published this and can that be tied to an entity
elsewhere.
5. DISCOVERY. Can anything find the page in the first place: sitemap,
internal links, and whether the page is linked from anywhere at all.
6. OFF SITE. What other pages say about me, which is the slowest and least
controllable stage and belongs last for that reason alone.
For each stage give exactly this:
- GATE: the condition that must be true before the next stage is worth any
time. Write it as something I can check in under five minutes, with the
exact thing to look at.
- STATE: PASS, FAIL or UNKNOWN, from what I pasted. Write UNKNOWN whenever I
have not given you the evidence. Never infer a PASS.
- DO: the actions for this stage, in order, each with a realistic time in
minutes. Only include actions that fit in the hours I told you I have.
- STOP RULE: what tells me this stage is done well enough to leave. Sized to
the morning, not to a perfect site.
Then a final section, THE MORNING. A single ordered sequence of the actions
that fit inside my available hours, drawn from the stages above, cut off
exactly where the time runs out. Say plainly what did not make the cut and
which stage it belonged to. Do not compress six hours of work into three by
shortening the estimates.
Constraints:
- Do not promise a ranking, a citation or a position. No operator publishes a
ranking system for any assistant, so there is nothing to promise.
- Mark anything not published by an operator as inferred, in the sentence
where you say it. Do not put inferred advice and documented behaviour in the
same voice.
- Do not invent statistics, timelines or expected results. If I ask how long
it will take to see an effect, say that there is no index status to check
and no published timeline.
- If a stage is UNKNOWN, the DO list for it is "go and measure this", nothing
else. Do not plan work on evidence I have not given you.
My site and the one page I care about: [DOMAIN AND ONE URL]
Raw word count from curl on that page: [THE NUMBER, OR "not measured"]
What my robots.txt allows: [PASTE YOUR ROBOTS.TXT, OR "not checked"]
Hours I actually have: [A NUMBER OF HOURS]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[DOMAIN AND ONE URL]- One page, not the whole site. The plan is only useful when it is concrete, and a single URL forces the model to give you actions rather than principles. Pick the page you would most want quoted, which is usually a service or comparison page rather than the homepage.
[THE NUMBER, OR "not measured"]- The word count of the visible text in your raw HTML, before JavaScript. It is the difference between stage two passing and stage two being the only thing that matters. If you have not measured it, write "not measured" and the plan will correctly send you to measure it first rather than guessing.
[PASTE YOUR ROBOTS.TXT, OR "not checked"]- The actual file, pasted. It decides stage one, which gates everything else. Pasting it also lets the model spot the group precedence trap, where a per bot group makes the wildcard rules stop applying to that bot entirely.
[A NUMBER OF HOURS]- How long you genuinely have, this week, not the ideal. This is what makes the last section a plan instead of a wish list. Three honest hours produce a sequence you will finish. Twenty imagined hours produce a document you will not open again.
How to run it
- 01Measure two things before you open the prompt
Fetch your robots.txt and copy it, then run a plain curl on your chosen page and count the words of visible text in the response. Both take under two minutes and they decide stages one and two. Without them the plan comes back full of UNKNOWN, which is correct and not useful.
- 02Be honest about the hours
The number you type here is the only thing stopping the output being a twenty item list you will never start. Write the hours you will actually spend this week. The final section is cut to fit it, and the things that fall off the end are a real result rather than a failure.
- 03Run it and go straight to THE MORNING
The stage by stage detail is reference material. The last section is the plan. Read it first, and if the first item is "go and measure this", do that before reading anything else, because a plan built on UNKNOWN evidence is a plan for somebody else's site.
- 04Respect the gates even when a later stage looks more interesting
Rewriting paragraphs is more appealing than reading a robots.txt, which is exactly why most sites in this position have well written pages that no crawler is allowed to fetch. If stage one fails, nothing after it counts, and the fix is usually one line and one deploy.
- 05Do stage three with a real editing pass, not a find and replace
Structure means question shaped headings with a standalone answer of about forty words under each. It is the stage where the prompt can only tell you what to do and you have to write the sentences. Budget more of the morning for it than the estimate suggests.
- 06Score the page at the end and keep the number
Run the finished page through the free AI content readiness check and write the score down with the date. There is no index status and no ranking report available here, so a repeated measurement of the property you were editing is the only feedback loop that exists. Rerun it after the next stage.
Questions people ask
How do you do answer engine optimisation?
In this order: confirm a crawler is allowed to fetch the page, confirm there is readable text in the raw HTML before JavaScript runs, break the page into sections with a standalone answer under each heading, name an author and tie the site to an identifiable organisation, make the page discoverable through a sitemap and internal links, and only then work on what other sites say about you. Each stage is close to worthless until the one above it is true.
What should I do first if I only have an hour?
Check robots.txt and run one curl against your most important page. Those two measurements take minutes and they tell you whether your problem is access, an empty response, or something further down. An hour spent editing prose on a page no crawler can fetch is an hour spent on nothing.
Is there a tool that does answer engine optimisation for me?
Tools measure, they do not do it. A checker can tell you a crawler is blocked, that your raw HTML holds forty words, or that no section opens with an answer. Removing the block, changing how the page is rendered and rewriting the openings are all decisions with tradeoffs, and none of them is a toggle.
How long before answer engine optimisation shows a result?
Nobody can tell you honestly, because there is no index status to check and no operator publishes a timeline. What you can observe is your own measurements moving: crawler access, raw word count, whether sections open with an answer. Treat any specific number of weeks you are quoted as invented.
Do I need to know whether this is called AEO or GEO?
No, and the argument is a distraction from the work. Neither term has a standards body or a defining document behind it, so both labels sit on top of the same sequence. If a supplier is selling you one of them, the useful question is which stage of that sequence they are actually doing.
Unsubscribe in one click. We never pass your address on.