practice

Building the internal business case for AEO work

How to justify AEO and GEO work internally: frame the risk, promise readiness instead of citations, scope a pilot, and agree success measures in advance.

The business case for AEO work rests on one promise you can keep: that the pages which matter to the business will be reachable, readable, and quotable by answer engines, and that you can show it page by page. It does not rest on a promise of citations, traffic, or revenue from AI answers, because nobody on your side of the page controls those. A case built on the first promise survives its first review meeting. A case built on the second fails the first time someone asks for the number.

This post is the case itself, in the order you would put it on paper: the problem, the costs, what can be promised and what cannot, a pilot scope, and the measures you agree before the work starts. It belongs to Running AEO and GEO as a practice. If you are building the case for a client rather than an employer, scoping and pricing an AEO audit covers the commercial side.

What problem are you asking money to solve?

The problem is that a growing share of search sessions now end with a generated answer, and your pages either feed that answer or they do not. State it with evidence, not with forecasts.

The cleanest public evidence on behaviour is from Pew Research Center, which analysed the browsing data of 900 U.S. adults during March 2025. Users who encountered an AI summary on Google clicked a traditional search result in 8% of visits; those who did not see one clicked a result in 15% of visits. Clicks on the links inside the AI summary itself occurred in 1% of visits to pages that had one (Pew Research Center, 22 July 2025). That is one study, one country, one month, and one engine, and it should be cited as exactly that. It is still enough to establish that the click is no longer the only outcome of a search.

Then add the evidence from your own data. In the Search Console Performance report, look for pages where impressions are steady or rising while clicks fall. In analytics, check whether assistants already send you referral sessions. Neither proves cause, and the case should not claim it does. Together they show the question is live for your site, not just for the industry.

Google’s own guidance helps the framing. Its documentation on AI features states that “there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary,” and recommends ordinary practices such as making content findable through internal links, available in textual form, and described by structured data that matches the visible text (Google Search Central: AI features and your website). That is useful in a budget meeting. It means AEO work is not a speculative new channel. It is technical and editorial work on pages you already own, aimed at a specific set of failure modes.

What does the work cost?

Most of the cost is people’s time, and the case should break it into the three kinds of time it actually takes.

CostWho spends itWhat it buys
AuditSEO lead or consultantA list of readiness findings per template and per page
Template fixesDevelopersChanges that apply to every page on a template: structure, markup, rendering
Copy fixesWriters or subject expertsRewritten section openings, definitions, sourced claims
VerificationSEO leadA re-crawl that shows each finding is gone

Template fixes are where the leverage is. One ticket that moves an image below a section’s opening paragraph changes thousands of product pages at once. Copy fixes scale linearly with the number of pages, so they are where scope has to be capped. Auditing answer-readiness across 10,000 URLs shows how an audit separates the two, and the split is the most useful number to put in the cost section: “two template tickets and roughly N sections of copy” is a cost a manager can approve.

Add the tooling cost honestly, including the time to run it. Then add the cost nobody writes down: the meeting time spent arguing about numbers that cannot be defended. A case that promises only measurable outcomes saves most of that.

What are the risks of doing nothing?

The risks worth naming are the ones you can check today, not projections of lost traffic.

  • Access risk. A robots.txt rule, a CDN setting, or a firewall may already stop search-oriented AI crawlers from reaching important pages. That is a binary fault with a one-line fix, and it is common enough to check first. How to check whether AI crawlers can actually reach you is the method.
  • Representation risk. If an answer engine does summarise your product, it works from whatever it could extract. Pricing locked in images, specifications behind scripts, and definitions buried in the fourth paragraph mean the summary is built from what was left.
  • Measurement risk. Without a baseline, any later conversation about AI search is anecdote. A crawl and a referral baseline taken now are what make next year’s report possible.

Each risk maps to a check, and each check produces a finding you can show. That is the property that separates a business case from a trend deck.

What can you promise, and what can’t you?

The section that decides whether the case survives is the one that draws this line. Put it in a table so nobody can miss it.

You can promiseYou cannot promise
Priority pages are reachable by the AI crawlers you choose to allowThat any assistant will cite those pages
Priority pages answer their headings in the first sentenceA share of voice in AI answers
Structured data on priority templates parses and matches visible contentA ranking or traffic increase from that markup
Every fix is verified by a re-crawl, with a before-and-afterA revenue figure attributable to AI answers
A monthly report of assistant referrals and AI-surface impressions, as reportedThat those figures will rise in a given quarter

The left column is readiness: deterministic facts about your own pages. The right column is outcomes: probabilistic, multi-factor, and decided by systems you do not run. What GEOBee measures, and what it deliberately doesn’t makes the same distinction for tooling. The reason to state it in the business case is defensive. If the case promises outcomes, the project is judged on outcomes, and a quarter with flat AI referrals ends it, however much readiness improved.

If someone asks why you cannot promise citations, the honest answer is short. Citation depends on the question asked, the competing sources, and the engine’s own retrieval and judgement. Measurement of it is noisy too; what prompt tracking can and can’t tell you covers why the same prompt can return different answers on different runs.

What should the pilot cover?

A pilot should be small enough to finish in one quarter and specific enough to produce a before-and-after. Scope it by template and by business value, not by a URL count.

  1. Pick two or three templates that carry money or leads. Pricing, the main product template, and the documentation template are typical. Exclude templates that do no job for the business.
  2. Fix access first. Check robots.txt and the edge for the crawlers you want to allow. This is the cheapest and most binary part of the work.
  3. Ship template fixes before copy fixes. They reach the most pages per hour of effort.
  4. Cap the copy work. Rewrite section openings on the top pages by clicks within the pilot templates, to a number the writers can finish.
  5. Re-crawl and compare. Close each finding only when a fresh crawl shows it is gone.

Prioritizing AEO fixes gives the scoring sheet for deciding which findings make the pilot. The pilot’s purpose is to prove the loop works on your stack: that findings can be found, fixed, shipped, and verified at a predictable cost. That is what earns the second quarter.

Which success measures should you agree up front?

Agree the measures before the work starts, in writing, and split them into two groups that are never blended.

Readiness measures, which the project owns. The share of priority pages with no Critical findings. The number of templates where the first element under each heading is a text answer. The number of priority pages reachable by each allowed AI crawler. Each is computed from a crawl, so it can be recomputed by anyone and compared month to month.

Outcome indicators, which the project reports but does not own. Assistant referral sessions in analytics. AI-surface impressions in Search Console, where the property has the report. A manual log of a small set of questions run on a schedule, recorded as a range rather than a single result. Measuring AI visibility without overstating it sets out how to collect each without inflating it.

Write down in the case that the first group is the pass/fail criterion for the pilot, and the second group is context. If the outcome indicators move, report them. If they do not, the pilot still succeeded when the readiness measures did, and the case said so in advance.

A one-page version for the decision-maker

Most business cases are read as one page. The fictional example below shows the shape.

Problem. Answer engines increasingly summarise searches in place of a click (Pew Research Center, 2025). We have no baseline for how our pages fare. Proposal. A one-quarter pilot on the pricing, product, and docs templates: fix crawler access, ship template-level structure fixes, rewrite section openings on the top 50 pages by clicks. Cost. Audit: 3 days of SEO time. Template fixes: 2 developer tickets. Copy: 50 section rewrites. Tooling: one license. Success. Pilot templates have zero Critical findings at re-crawl; every heading on pilot templates opens with a text answer; all allowed AI crawlers reach every pilot page. Reported, not promised. Assistant referrals and AI-surface impressions, monthly, from the start of the pilot. Not claimed. Citations, rankings, or revenue from AI answers.

The figures in that example are invented. Replace them with yours, and keep the last line.


To produce the baseline crawl and the before-and-after comparison the pilot needs, download GEOBee. The 7-day trial is enough to run the first audit on your pilot templates.

On this page