AEO agency SOPs are the written procedures that make answer engine optimization repeatable: how a prompt set is built, how answers are sampled, how an answer-first page is checked, and how results are reported. They are separate from SEO SOPs because the input is a question instead of a keyword, and the check is a sampled answer instead of a ranking. This post lists the SOPs to write, then writes three out as checklists.
The field structure for any SOP is in SEO agency SOPs: purpose and trigger, owner, inputs, steps, checks, output and review date. Use the same structure here. What changes is the content, because AEO has failure modes SEO does not, mainly answers that vary from run to run. The operating model these procedures sit inside is in AEO agency operations.
The AEO agency SOPs to write
An AEO service needs procedures for the tasks that are new to an SEO team, plus adjusted versions of the ones it already has.
- Prompt set build. How questions are collected, grouped by intent and approved.
- Baseline sampling. The first measurement run, before any change.
- Citation sampling run. The repeating measurement, with fixed prompts, engines and run counts.
- Answer accuracy review. A human check of what engines say about the client.
- Answer-first page brief. What a page must specify before it is written.
- Answer page QA. Checks run before an answer-first page ships.
- Structured data validation. How markup is specified, tested and released.
- AEO report. How sampled results become a client report.
Intake, access and scope-change SOPs from your SEO set usually carry over with small edits, mainly adding prompt groups to the scope document.
Worked SOP 1: Prompt set build
Trigger: AEO scope signed. Owner: AEO lead. Inputs: client interview notes, sales or support questions, the client's category and competitors. Output: an approved prompt set, filed with a version date.
- Ask the client for the questions buyers ask before, during and after a purchase decision. Request real wording from sales calls or support tickets where available.
- Draft prompts in natural language, as a person would type them to an answer engine.
- Group prompts by intent: category, comparison, problem, and brand-specific.
- Remove duplicates that differ only in wording, but keep two phrasings for the most important prompts.
- Mark which answer engines matter to this client and note the reason.
- Mark any location or language the prompts must be tested under.
- Send the set to the client for approval. Record who approved it and the date.
- Freeze the set. Changes go through a new version, never an edit in place, so trend lines stay comparable.
Checks: a second team member confirms every prompt is phrased as a question a buyer would ask and not as an internal term. The set does not go into sampling until the client approval is recorded.
Worked SOP 2: Citation sampling run
Trigger: the sampling date in the client calendar. Owner: AEO analyst. Inputs: the current approved prompt set, the agreed engine list, the recording sheet. Output: a completed sampling record and a one-paragraph read of changes.
- Confirm you are using the current frozen prompt set version.
- For each prompt and each engine, run the prompt the agreed number of times in a clean session.
- For each run, record the date, engine, and whether the answer cites any of the client's pages.
- Record whether the brand is named in the answer text, even if no link appears.
- Record which other sources and brands appear.
- Save the full answer text or a screenshot for any run where the client is cited or named inaccurately.
- Compute the rate of citation and the rate of mention per prompt group, not per single run.
- Compare with baseline and the last run. Mark differences that are within normal run-to-run variation as unchanged.
- Write a short read: what changed, what did not, what to check next.
Checks: the analyst does not report a prompt as won or lost from one run. A reviewer confirms the numbers match the recording sheet and that variation is acknowledged in the read. The limits of the method are stated in every record: answers vary, and tooling for this is still maturing.
Worked SOP 3: Answer page QA
Trigger: an answer-first page is marked ready. Owner: QA reviewer, not the writer. Inputs: the brief, the draft, the prompt group it serves. Output: a signed QA record.
| Check | Passes when |
|---|---|
| Opening | The first two sentences answer the target question without the rest of the page |
| Specificity | The answer states something concrete; a vague opening fails |
| Claims | Each factual claim has a source, a date or a named basis a reader can verify |
| Brand facts | Names, locations, offerings and claims match the brand fact sheet |
| Headings | Phrased as the questions or topics buyers use |
| Structured data | Validates and matches what is visible on the page |
| Crawl and links | The page is crawlable, indexable and linked from relevant pages |
| Attribution | Author or organization attribution is present where the brief requires it |
| Record | Pass or fail is logged per line and failures go back with the exact line |
| After release | The page is added to the next sampling run for its prompt group |
Checks: sign-off carries reviewer name and date. A page with an unverified claim does not ship.
Where AEO SOPs go wrong
- Editing the prompt set mid-stream. Trend lines become meaningless. Version it.
- Reporting one run. Answers vary. Report rates across runs.
- Writing steps that depend on a specific tool screen. Tools change. Describe the action and the record.
- Skipping the post-release link. Shipped pages never enter sampling, so nobody learns whether they worked.
A missing SOP often shows up as hours nobody can account for, which affects the utilization number I explain in the benchmark guide. Mapping these procedures against your team's capacity is the work behind an AEO agency utilization consultant engagement. The concepts behind these procedures are in the AI search optimization guide, and the entity-side counterpart is GEO agency SOPs.
FAQ
What SOPs does an AEO service need?
At minimum: prompt set build, baseline sampling, a repeating citation sampling run, answer accuracy review, answer-first page brief, answer page QA, structured data validation and an AEO report. Intake and scope-change SOPs from your SEO set carry over with small edits.
How is an AEO SOP different from an SEO SOP?
The structure is the same, but the inputs and checks differ. AEO SOPs start from a frozen prompt set, measure by sampling answers repeatedly across engines, and QA pages by whether the opening sentence answers a question on its own. SEO SOPs start from keywords and check rankings and technical correctness.
How often should an agency sample AI answers?
Set a fixed schedule per client and keep it, such as monthly, with enough runs per prompt to show variation. The right number depends on the client and the engines involved. What matters is consistency, because a changing method makes trend comparisons unreliable.
Why version the prompt set?
Because editing prompts in place breaks comparisons over time. A new version starts a new baseline, and the old data stays interpretable. Record who approved each version and when, so a change in results is never confused with a change in the questions.