Why answer engine optimisation matters, and when it honestly does not
The business case for AEO without the invented statistics, when ignoring it is the correct call, and a prompt that writes you a one page decision memo.
- Works in
- ChatGPT, Claude, Gemini
- You need
- How your customers found you in the last ten deals · What one customer is worth and how long they take to close · What an assistant says about you today, if you have asked
- Written for
- why is answer engine optimization important
Scored by our own engine
This page, run through the audit we sell. Measured 4 August 2026.
Answer engine optimisation matters for one structural reason. When a buyer gets an answer instead of ten links, there is one description of you rather than a page of results to compare, and if nothing of yours can be fetched or quoted, that description is assembled from whatever else the web says. The prompt below turns that into a decision memo, including the version where the correct answer is to do nothing.
The case, without the invented numbers
Every figure you have read about the share of searches now answered by AI was estimated by somebody selling something. We will not repeat any of them.
What can be said without a number is the shape of the change. A list of results is a comparison, and a generated answer is a verdict. In a comparison, being fourth still gets you read. In a verdict, being absent means being absent, and being described inaccurately is worse than being absent because it is confidently wrong at the moment of decision.
The second reason is cost. Most of the work overlaps with things worth doing anyway: letting crawlers fetch you, putting your content in the raw HTML, answering the question in the first sentence under the heading that asks it. The overlapping work carries no bet. It pays for itself on classic search and on readers whether or not any of this accelerates.
The case for ignoring it
There is a real one, and it depends entirely on how your customers find you.
If your last ten deals came from referrals, a procurement framework or a trade body, then assistant visibility is a lottery ticket and the money belongs in the channel that is working. If you are a publisher whose revenue depends on the visit, being summarised well and never visited is a loss. If nothing on your site is crawlable yet, the prior work is bigger than this argument.
And there is no reporting. No console counts assistant citations, answers vary between sessions, and spending is hard to govern in a category with no measurement. That is a legitimate reason to wait, and not one anybody selling AEO will give you.
Why the prompt argues against itself
Ask a model whether you should invest in AEO and it says yes, because the question was leading and yes is the agreeable answer.
This one has to pick one of three verdicts, justify each reason against something you told it, and then write the strongest honest argument for the opposite decision as if trying to win. If the counter argument reads as a caveat, the memo has not done its job.
Start with the definition it rests on in what answer engine optimisation means, settle the vocabulary at AEO and GEO, which is which, and before any of it, check whether AI crawlers can reach you at all. If the verdict turns on budget, what AI visibility tools cost is the price side of the same decision.
You are advising a business owner on whether to spend money on answer engine
optimisation. Write a one page decision memo. Not an explainer, not a pitch,
and not a list of tactics.
Open with a verdict, one of exactly three, in the first line:
INVEST NOW. INVEST PARTIALLY, meaning do only the work that is worth doing on
other grounds. DO NOT INVEST YET.
Choose it from what I tell you below, not from a default. DO NOT INVEST YET is
a real answer and you should reach it when the evidence points there.
Then, in this order:
THE CASE, three reasons specific to my business, not to businesses generally.
Each one has to name something I told you. A reason that would apply to any
company is not a reason, it is filler, and I will notice.
THE COST. What the work actually is, in hours and in who does it, split into
work on my own site and work on other people sites. State plainly which of the
two is slower and less controllable.
THE CASE AGAINST YOUR OWN VERDICT. The strongest honest argument for the
opposite decision, written as if by somebody who wants to win. Not a token
caveat. If you recommended investing, argue that the money belongs elsewhere
and say where. If you recommended waiting, argue what I lose by waiting.
WHAT WOULD CHANGE THE VERDICT. Two or three observable events that should make
me revisit this, each specific enough that I would recognise it when it
happens. Not "as AI adoption grows".
WHAT I WILL NOT BE ABLE TO MEASURE. Be blunt. Say which effects have no
reporting behind them, that assistant answers vary between sessions, and what
the weakest but genuine evidence looks like: asking the same question from
fresh sessions over weeks and watching whether the answer changes.
Constraints, and these are the whole point:
- Do not invent statistics. No share of searches, no traffic decline figures,
no adoption percentages, no citation share. You do not have those numbers
and neither does anybody quoting them at me.
- Do not claim answer engine optimisation is a standard or that any body
defines it. It is an industry coinage.
- Do not promise a ranking, a citation or a position in any assistant.
- If what I told you is too thin to reach a verdict, say what one piece of
information you need instead of guessing.
What I sell and to whom: [ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]
How my last ten customers found me: [LIST THE CHANNELS, ROUGH COUNTS ARE FINE]
What a customer is worth and how long they take to close: [VALUE AND SALES CYCLE]
What an assistant says about me today: [PASTE THE REPLY, OR WRITE "not asked"]What to change
Everything in square brackets is yours to replace. Nothing else needs editing.
[ONE SENTENCE ON WHAT YOU SELL AND TO WHOM]- The buyer has to be in the sentence. "We sell compliance software to UK insurance brokers" produces a memo about a researched purchase. "We sell compliance software" produces a memo about nobody, and every reason in it will be filler.
[LIST THE CHANNELS, ROUGH COUNTS ARE FINE]- The single most important input, and the one that decides the verdict. "Six from referrals, three from a trade body directory, one from Google" is a business where the memo should probably tell you to wait. "Eight from organic search" is not. Count the last ten real deals rather than guessing at a percentage.
[VALUE AND SALES CYCLE]- Roughly what a customer is worth over their life with you, and how long from first contact to signature. This is what turns the cost section from hours into a decision. A long cycle with a high value means being named once in an assistant is worth a great deal; a low value impulse purchase usually means it is not.
[PASTE THE REPLY, OR WRITE "not asked"]- Ask an assistant to recommend suppliers in your category and paste the whole answer. It is the only evidence in the run. Being described inaccurately is a different problem from being absent, and the memo reaches a different verdict for each.
How to run it
- 01Count your last ten deals before you write anything
Open the CRM or the invoices and write down how each of the last ten customers actually reached you. Not the attribution report, the real story. This single list decides the verdict, and a business that closes on referrals has a very different answer from one that closes on search.
- 02Ask an assistant to recommend suppliers in your category
Use a fresh or temporary chat with search enabled so history does not colour the reply, and ask the question a buyer would ask. Copy the whole answer with its citations. Being named wrongly, being absent, and being named accurately are three different starting positions.
- 03Run the memo and read the verdict last
Read the case, the cost and the counter argument before you look at which of the three it chose. The reasoning is the useful part. A verdict you agree with for reasons that are filler is worth less than a verdict you disagree with for reasons that are specific.
- 04Check whether the case against is real
The counter argument is where models go soft. If it reads as a caveat rather than as somebody trying to win, ask it again to argue the opposite decision properly. A memo that cannot be argued against was not a decision, it was a recommendation with a decoration on it.
- 05If the verdict is partial, do only the overlapping work
Unblocking crawlers, putting content into the raw HTML, and writing an answer under each heading are all worth doing for classic search and for readers regardless. That is the work with no bet in it, and it is the whole of a partial investment.
- 06Establish the baseline you will be judged against
Before changing anything, run the free AI crawler check and save the result, and save the assistant answer you collected. Neither is reported anywhere, so if you do not record where you started, you will not be able to tell later whether anything moved.
Questions people ask
Why is answer engine optimisation important?
Because a growing share of buying research now ends in an answer rather than a list of links, and in that format there is one description of you rather than ten results to compare. If your page cannot be read or quoted, the description a buyer gets is assembled from whatever else exists about you. How large that share is nobody can tell you honestly, which is why the case has to be made from your own customers rather than from an industry figure.
When is it reasonable to ignore AEO entirely?
When your customers do not research this way. Businesses that close on referral, on procurement frameworks, on trade relationships or on walk in trade have a defensible case for spending the money elsewhere. So does a publisher whose revenue depends on the visit itself, since being summarised well and never visited is a loss rather than a win.
Is AEO worth it for a small local business?
Usually yes, and mostly not on your own website. For a local business the sources an assistant reads are disproportionately third party: maps listings, review platforms, directories and local roundups. Getting the name, address, category and opening hours identical across those is cheap, it helps ordinary local search at the same time, and it does more than rewriting your homepage.
How do I measure whether any of this worked?
Poorly, and honestly poorly is still better than not at all. There is no console reporting assistant citations, and answers vary between sessions and accounts. The available method is to fix a set of buyer questions, ask them from fresh sessions on a schedule, and record which sources are named. Look for a change that holds over weeks rather than reading a single reply as a result.
Is answer engine optimisation just a rebranded SEO retainer?
Often, and the test is simple. Ask what the supplier is doing about the sources other than your website, and ask which deliverables would be different if the label had never existed. Most of the work is genuinely shared with classic SEO, which is not fraud. Charging a premium for the shared half and doing none of the unshared half is.
Unsubscribe in one click. We never pass your address on.