What should you fix before you hire an AEO agency?

Fix the four things that are cheap for you and expensive for an agency to buy: crawler access, server-rendered text, your real buyer question list, and a named internal approver. Nothing else on the usual pre-engagement checklist is worth your time, and a competent agency can do the rest faster than you can.

The logic is about where retainer hours go rather than about capability. An agency billing monthly will spend its first weeks establishing facts, and two of those facts are things only you can supply. If the agency has to reverse-engineer your buyer questions from keyword tools, you are paying specialist rates for a worse version of a list your sales team could write in an hour.

Crawler access and rendering sit in the same category for a different reason. Both are binary gates, both are usually one afternoon of developer work, and until they pass, nothing anybody does can be measured. An engagement that starts against a site the engines cannot read produces a month of activity and no signal.

What is not worth doing yourself is the content work. Rewriting pages before the baseline is measured destroys the comparison the engagement depends on, and page structure is the part an agency is actually better at. The split is simple: you supply access and knowledge, they supply the writing and the measurement. Our page on what an AEO engagement includes sets out which side each deliverable usually falls on.

Which of these can you actually fix yourself in a week?

Three of the four fix in a week, and the fourth is a conversation rather than a task. The table sets out who does each one and why it matters, so you can hand it to a developer and a sales lead without explaining the category first.

What to fix	Who does it	Why it has to happen first	Time
Crawler access in robots.txt and at the CDN	Whoever owns the CDN, not marketing	A blocked crawler is a retrieval failure. No content work can reach past it, and bot-management rules at the CDN often block more than the site owner realises	An afternoon
Text that renders without JavaScript	Front-end developer	A page whose body appears only after scripts run can be fetched and still read as empty	A day to diagnose, longer to fix if the stack is client-rendered
The ten questions buyers ask in first calls	Two salespeople, in their own words	Nobody outside your company has this list, and it is the input that decides what gets written	An hour, and it is the highest-value hour in the whole process
One named person who can approve copy	You, before the kick-off call	Approval latency is the most common reason an engagement produces three pages instead of twelve	A decision, not a task

Run the first two before you take a single sales call, because their answers change what you should be buying. A site that cannot be fetched needs a developer, not a retainer. Our guide to checking whether AI can read your website covers the exact fetches to run, and whether AI crawlers read JavaScript covers the rendering half.

Why is schema markup the wrong thing to fix first?

Schema markup is the wrong first fix because the one test of it with matched controls found no uplift in AI citations. Almost every pre-engagement checklist published by an agency puts structured data near the top, and the measured evidence does not support that position.

Ahrefs tracked 1,885 pages that added JSON-LD between August 2025 and March 2026, matched each against control pages with similar prior citation levels that never added it, and ran a difference-in-differences comparison. ChatGPT came out at +2.2% and Google AI Mode at +2.4%, both statistically indistinguishable from zero, and AI Overviews at -4.6%, which was statistically real and which the authors say they cannot attribute to schema alone. Their summary is that adding schema produced no major uplift on any platform.

Read the limits before anyone quotes that as "schema does not work", because it is narrower than the headline. Every page in the treated set already had more than 100 AI Overview citations, so the result describes pages that already get cited rather than pages that do not. Schema types were pooled although they do different jobs, the window was 30 days, and schema changes often arrived alongside other edits. The accurate version is that on pages already being cited, adding schema did not measurably increase citations inside the window measured.

Google's own position points the same way. Its documentation on AI features and your website states there are no additional requirements to appear in AI Overviews or AI Mode and no special optimisations necessary, and that structured data should match the page's visible text. So keep your schema valid as hygiene and as a parsing aid, and spend the first week on the four items above instead. Our page on whether schema markup helps AI citations goes through the argument in full.

What does an AEO agency need from you that money cannot buy?

An agency cannot buy the questions your buyers ask in the words they use, and that list is the single most valuable thing you can hand over. Everything else in the engagement is downstream of it, because a page targeting the wrong phrasing is a well-built page aimed at nobody.

The evidence that this is the decisive input is unusually direct. Chen and Koudas, at the University of Toronto, analysed 158,099 answers and 468,037 citations across two 90-day collections using a design that compares pages against other pages on the same site. Within a single site, the pages cited more often were the ones whose titles and paragraphs matched the questions people actually ask. FAQ blocks, structured data, summary boxes and page speed showed effects only when comparing across different websites, which means they travel with the kind of site that has them rather than predicting which page on a site gets cited.

The same study is worth reading in both directions, which is how we present it to clients. Page properties explained about 2.8% of the variation between pages on one site, against about 32% for which pages the engine's own retrieval returned. That is adverse to anyone claiming a page score predicts how often you will be cited, including anyone selling page work. What it supports is narrower and real: the one page-level property that separates two pages on the same site is whether the language matches the question.

Get the list from people who hear the questions. Ask two salespeople what comes up in every first call, read the last fifty support tickets, and check the lost-deal notes for the objection nobody has a page for. Write them down as sentences, not keywords, because that is how they arrive at an assistant.

What should you deliberately leave alone before the agency starts?

Leave your existing pages alone, including the ones you are embarrassed by. Reorganising the site in the weeks before an engagement starts is the most expensive form of helpfulness available, and it costs you the only clean comparison you will get.

Two things break when you tidy up first. The baseline stops meaning anything, because a measurement taken after a restructure cannot separate the agency's effect from your own, and that comparison is what the renewal conversation will turn on in six months. And the pages you are most tempted to delete are often the ones the engines already fetch, which makes them the cheapest wins on the whole site. Rewriting a page that is already retrieved is faster than earning retrieval for a new one, and deleting it throws away work you already have.

Resist consolidation for the same reason. Merging two thin pages into one strong page is often correct, but which two is a question the measurement answers, and doing it on instinct in week zero usually merges the page that was working into the page that was not. Our page on whether two of your pages can compete for the same AI answer covers how that call should be made.

One exception is worth naming. Fix anything factually wrong about your company on your own pages, because that is not a measurement question and it propagates. Everything else can wait three weeks, and waiting is cheaper than it looks.

How do you tell whether you are ready to hire at all?

You are ready to hire when assistants already produce vendor answers in your category, your pages can be read without JavaScript, and one named person can approve copy inside a week. Missing any of the three means a retainer would be paying for something other than the work.

Test the first one yourself, without buying anything. Put your ten buyer questions to ChatGPT, Gemini, Perplexity and Google AI Mode without naming your company, and see whether the answers recommend specific suppliers or stay generic. A category where five competitors get named has a contest worth entering, and whether AEO works when nobody searches your category covers what to do when it does not.

The third condition is the one companies get wrong most often, and it is not a capability problem. An engagement that produces twelve pages a quarter needs copy approved continuously, and a review chain that routes every page through legal and two executives will produce three. Decide before you sign who holds the pen, what they can approve without escalating, and what the turnaround is.

Where this adds up to "not yet", the honest sequence is a technical fix and a prompt test rather than a retainer. Where it adds up to "ready", the first month should be measurement rather than publishing, and our page on what an AI visibility audit includes covers what that month should produce. A free visibility assessment will tell you which of the three conditions you are missing before you commit to anything.

What else do people ask about preparing for an AEO engagement?

The five questions below come up in assessment calls with teams about to put a retainer through procurement.