How do you do AEO when legal and compliance have to approve every page?
Compliance is rarely what stops AEO in a regulated company. The review queue is. Most of what moves AI citations on an existing page is rewording and reordering what a reviewer already cleared, and the measured within-site levers are exactly that: Chen and Koudas, University of Toronto, October 2026, found that, inside a single website, a page's title match to the question asked was worth +14.1% citations per standard deviation and its best-paragraph match +15.5%, while FAQ blocks, schema markup and page speed showed no within-site effect at all. Those reworded headings almost never create a new regulated claim. What does cost you is cycle time, because the same study found that only 14% of the repeatable difference between a site's pages was still shared between two sampling windows ten weeks apart on ChatGPT, and 29% on Gemini. So separate the claim work from the structure work, batch the claims with their evidence, and ask legal to approve a pattern once rather than every page.
How do you do AEO when legal and compliance have to approve every page?
You do AEO under legal and compliance review by splitting the work into changes that move words a reviewer already cleared and changes that make a new factual claim. The first group ships on an editorial cadence and carries most of the citation effect. The second group goes through the full path, batched, with its evidence attached.
The review requirement is not something an agency gets to design around, and anyone who offers to is telling you something useful about themselves. A broker-dealer publishing a retail communication has to have it approved by an appropriately qualified registered principal before the earlier of its use or its filing, and the same rule sets filing windows of 10 business days before or after first use depending on the category, per FINRA Rule 2210, read on finra.org on 22 October 2026. Medical, legal and insurance review regimes impose their own versions of the same constraint. Cycle time is a fixed input to the plan.
What changes the outcome is how much of the programme is forced through that queue. A supplier who treats every page as a fresh draft puts the whole backlog in front of your reviewers, then reports slow progress as your problem. A supplier who knows which changes carry the citation effect puts a small number of claim decisions in front of them and does the rest as restructuring.
That split is also the honest way to scope the engagement. Ask for the work to be quoted as two streams, one that depends on your review calendar and one that does not, and ask what the second stream alone is expected to move. If the answer is nothing, you are buying content production rather than visibility work.
Which AEO changes usually clear review without a new claim?
The changes that clear review fastest are the ones that reuse approved sentences in a different order: rewriting an H2 into the question a buyer actually asks, moving the answer to the first sentence of a section, splitting a 900-word block into three self-contained ones, and building an FAQ from wording that has already been signed off.
None of that is cosmetic, which is the part most review-bound teams get wrong. Chen and Koudas, University of Toronto, October 2026, ran a within-website design over two 90-day collections, 143,249 answers and 396,188 citations in US home internet plus 14,850 answers and 71,849 citations in a niche business market, collected between 24 June and 21 September 2026. Comparing pages on the same site, the only features that passed all five of their robustness tests were title match to the question, worth +14.1% citations per standard deviation (95% confidence interval 10.7 to 17.6), and best-paragraph match, worth +15.5% (11.1 to 20.1).
The same paper found that the markup work most enterprise programmes lead with does not separate pages on one site. FAQ blocks, summary boxes, author boxes, schema.org markup and page speed all looked positive when compared across websites, and inside a single website the estimates were -3.4% for FAQ blocks, -5.3% for schema.org markup and -1.1% for page speed. The authors attribute the gap to Simpson's paradox: those features cluster on sites that are cited more for other reasons. They describe their results as associations, not causal effects, and that caveat belongs in any planning meeting that quotes them.
Read together, that is good news for a regulated team. The lever that survives the strictest test available is the wording of a heading and a paragraph, and rewording a heading into a question rarely creates a claim anyone has to clear.
What does a compliance-safe change look like next to one that needs sign-off?
A compliance-safe change moves or reshapes approved copy. A change that needs sign-off introduces a fact, a number, a comparison or a third party. Sorting every item on the backlog into those two columns before anyone starts is what keeps the queue short, so the table below is the one worth agreeing with your reviewers in the first week.
| Change | What it touches | Does it add a claim? | Usual path |
|---|---|---|---|
| Reword an H2 into the buyer's question | Heading wording only | No | Editorial |
| Move the answer into the first sentence | Order of approved sentences | No | Editorial |
| Split long sections, add a summary box | Structure, with copy reused verbatim | No | Editorial |
| Build an FAQ from cleared sentences | Repetition of approved wording | No, if the wording is unchanged | Editorial, with a check that nothing was paraphrased |
| Add FAQPage or Article markup | A machine-readable copy of visible text | No, provided it matches the page word for word | Technical, and in scope for review wherever metadata is read |
| Add a statistic, benchmark or outcome figure | A new factual claim | Yes | Full legal and compliance path, evidence attached |
| Name a customer, result or testimonial | A claim about a third party | Yes | Full path, plus written consent |
One row on that table causes more arguments than the rest. Structured data is a copy of your claims in a form a machine reads first, so a page whose markup says something the visible copy does not is a review failure waiting to be found. Our own rule is that the FAQ markup has to match the visible answers word for word, which also happens to be what keeps the page honest when an engine quotes it.
How much does a slow approval cycle cost you in AI visibility?
A slow cycle costs you the window in which a citation is live, and that window is shorter than most annual plans assume. Citations turn over continuously, so a page sitting in review is not holding its place in a queue. It is absent from a set that is being reshuffled while it waits.
Chen and Koudas, University of Toronto, October 2026, measured the turnover directly. Across their 90-day collections, about 75% of the variation in citation counts within a website was repeatable but explained by no page feature they tested, and that repeatable part decayed as the sampling windows moved apart: two windows two weeks apart shared 40% of it on ChatGPT, and ten weeks apart only 14%, against 29% on Gemini and none on Perplexity.
Day-to-day churn is larger still. Schulte, Bleeker and Kaufmann tracked ChatGPT, Gemini, Google AI Mode and Perplexity across four verticals for 45 to 46 days between 24 January and 20 March 2026, comparing 4,044 consecutive-day pairs, and found day-to-day overlap of cited sources averaging a Jaccard similarity of 0.34 to 0.42, which they read as roughly 35% of sources surviving from one day to the next.
Two practical consequences follow. First, a six-week review queue means the page you approved was written against an answer set that has since largely turned over, so publishing order should follow buyer value rather than whatever happens to clear first. Second, nobody should report a single reading of a prompt as a result, yours or your agency's. We baseline the same prompt set twice, a week apart, before changing anything, precisely so the first number you see includes your own noise. The arithmetic behind that cadence is in our guide to measuring AI citation decay.
Should you ask legal to approve a pattern instead of every page?
Ask for a pattern approval. The reviewable object is not the page, it is the set of claims the page is allowed to make, and once that set is agreed the structural work can run without a queue. Three artefacts carry most of it, and all three are things your reviewers are used to producing.
The first is a claim library: the approved wording for every factual statement your pages may repeat, with the evidence and the approval date beside each entry. The second is a tiered content model, so a definitional page, an FAQ and a performance claim travel different paths rather than all waiting on the slowest one. The third is a named owner for the question list, because the questions buyers ask are a commercial document and marketing should own which ones are worth a page.
What a pattern cannot do is cover a claim nobody has approved yet. If the strongest thing you could say about your product has never cleared review, that is the item to put in front of a reviewer first, and it is worth more than twenty structural fixes. The honest version of this work says so rather than quietly publishing the weaker page.
Our own engagements are sequenced around that order: score the pages, separate the structural backlog from the claim backlog, ship the structure, and run the claim decisions as a short list with evidence attached. Where the structural work alone does the job, it shows up in the readiness score, as in the two anonymised early case studies where one client's audit score moved from 17 to 68 after implementation. That figure is an input measure, not a citation count, and the same page says so.
What should you ask an AEO agency if you are in a regulated industry?
Ask an agency how it works when your industry's review cycle is slow, and listen for whether the answer is a process or a complaint. Six questions separate suppliers who have done this from suppliers who are about to discover your review calendar in month two.
Who writes the claim, and who holds the evidence behind it? A supplier who expects to insert statistics they sourced themselves is handing you an obligation you cannot discharge, because your reviewer will ask for the primary source and you will not have it. Will they work from your claim library and keep FAQ markup identical to the approved visible copy? Those two answers tell you whether the FAQ work will survive an audit.
How do they handle a page whose strongest claim is stuck in review? The useful answer is that the page ships without it and is revisited, not that the claim is softened into something vague enough to pass. Vague is its own compliance problem and it costs you the alignment that the measured evidence says matters.
What do they report when a month produces no publishing? Range-based reporting against a baseline is the honest answer, because one prompt run cannot separate your change from the engines reshuffling their sources. And will they accept your review service level in the plan rather than in the excuses? A plan built around a two-week legal turnaround is a different plan from one built around five days, and both are deliverable if nobody pretends otherwise.
If you want that mapped against your own site before you brief anyone, our free visibility assessment reports which of your pages are losing citations to wording rather than to claims. The wider question of what to settle before you hire is in what to fix before hiring an AEO agency.
What else do people ask about AEO in regulated industries?
The five questions below come up in assessment calls with regulated marketing teams whose copy goes through legal, compliance or medical review.
Does AEO require publishing new claims about your product?
AEO does not require new claims. The page-level feature that survives the strictest published test is whether a page's title and best paragraph match the question a buyer asks, which is a wording and ordering change rather than a factual one. New claims help where they are the strongest thing you could say, but a regulated team can run most of the structural programme on copy that has already been approved.
How long does legal review usually add to an AEO programme?
Review adds whatever your own cycle is, and the cost of that delay is measurable rather than theoretical. In Chen and Koudas's October 2026 study, two sampling windows ten weeks apart shared only 14% of the repeatable difference between a site's pages on ChatGPT, so a long queue consumes most of the window in which a page's advantage holds. The fix is to batch claim decisions and ship structural changes continuously, rather than to treat each page as one approval.
Can you add FAQ schema markup without compliance sign-off?
FAQ markup that repeats approved visible copy word for word adds no claim, so it usually travels as a technical change. The moment the markup paraphrases the page, it becomes a claim nobody reviewed, and it is the version an engine reads first. Keeping the markup identical to the visible answers is both the compliance answer and the citation answer.
What is a claim library, and why does it speed up AI visibility work?
A claim library is the approved wording for every factual statement your pages may repeat, with the evidence and the approval date beside each entry. It speeds the work up because writers stop drafting new claims that need review and start assembling cleared ones, which turns most of the backlog into an editorial task instead of a legal one.
Is AEO worth it for a company in a heavily regulated category?
It is worth it where buyers are already getting vendor answers from assistants in your category, and the review constraint changes the sequence rather than the verdict. Regulated categories often have thinner competition in AI answers precisely because competitors publish slowly, so the structural half of the work, which needs no new claims, tends to go further than it would in a looser market.