How to Estimate AEO ROI Before You Commit Budget, With No Data Yet

What does a pre-spend AEO forecast actually estimate?

A pre-spend forecast estimates the expected value of holding a place in the answers AI assistants give your buyers, and sets it against what the program costs. You are pricing an asset before you buy it, so every term in the model is an estimate.

Finance funds R&D, brand campaigns and headcount on estimates. What it will not fund is a request with no unit of account, which is how "we cannot measure AI citations yet" sounds.

Few finance teams dispute the channel itself. In a February 2024 press release, Gartner predicted that traditional search engine volume would drop 25% by 2026, with search marketing losing share to AI chatbots and other virtual agents.

A citation pays out in more ways than a click does, so name each one before the math:

• Captured demand. Buyers read your name in an answer, then click or search for you and arrive already shortlisted.
• Displaced competitors. Every slot you hold in an answer is one a competitor does not.
• Zero-click lift. Some buyers never click and turn up later through branded search or direct traffic.
• Durability. An earned citation keeps working until it is displaced; a paid click stops when the budget does.

The mirror image is the cost of a lost shortlist slot: your average deal value times your win rate, applied to the questions where a competitor is named and you are not.

What formula should the forecast use?

Use an expected-value chain in which finance can challenge each term on its own. Run it per cluster of close buyer questions, then add the clusters up.

Expected annual value = Annual question volume x AI research share x Capture rate x Act rate x Conversion rate x Average deal value x Gross margin x (1 minus confidence discount)

Forecast ROI is expected annual value minus the full annual program cost, divided by that cost.

Gross margin converts revenue into the profit finance counts; the table below defines every term.

Where does each input come from before you have data?

Every input has a source you can name today, even with no attribution history.

Capture rate is the only input your program moves directly.

How do you estimate how many of your buyers research with AI?

Take AI research share from a published survey of buyers like yours, use the lowest credible figure in your base case, and cite it in the model. Match the source to your market:

• B2B software. In G2's survey of B2B software buyers, 51% said they begin software research with an AI chatbot more often than with Google, up from 29% in April 2025.
• Consumer and e-commerce. Adobe's July 2026 consumer survey found 42% of consumers had used AI assistants for online shopping, according to Adobe's August 2026 AI traffic report.
• Buying decisions in general. McKinsey's 2025 survey, published in its report on AI search as the new front door to the internet, found 44 percent of AI-powered search users call it their primary and preferred source of insight, ahead of traditional search at 31 percent.

A survey counts people who use an assistant. It cannot tell you what share of searches for one question have moved to AI, so use the surveys to set a range, then add one question to demo forms and win/loss calls ("Did you use an AI assistant to build your shortlist?") to measure your own.

Why is keyword search volume a weak proxy for AI query volume?

Keyword volume counts what people type into a search engine, and AI assistants publish nothing comparable for individual questions. Use it as a labelled floor for question volume.

• Assistants publish totals only. OpenAI's study of how people use ChatGPT reported 700 million weekly active users of ChatGPT. It found about 49% of messages are "Asking", seeking information or advice, but neither figure says how often anyone asks about your category.
• Google blends AI traffic into search. Google's documentation on AI features says sites appearing in AI Overviews and AI Mode are included in the overall search traffic in Search Console's Performance report. Google's Search Console help page says a separate generative AI performance report was rolled out to all websites worldwide as of August 31, 2026.
• Long questions behave differently. Pew Research Center found only 8% of one- or two-word Google searches produced an AI summary, rising to 53% for searches of 10 words or more. A forecast built on short head terms measures the queries where AI answers appear least.

Build the question set from the full questions buyers ask, then take the volume of their nearest keywords.

How do you measure a baseline capture rate when you have none?

Run your buyer questions through the AI assistants and count how often your brand is named. That share is your baseline capture rate, and it may well be zero.

1. List the questions by persona. Different buyers get different answers, so we research each persona from your market (role, company type, goals, pain points and decision criteria).
2. Run each question more than once. On OnlyAEO's platform every prompt runs twice per engine, with ChatGPT, Claude and Gemini on by default with web search.
3. Record who is named and which pages are cited. McKinsey's analysis found a brand's own sites often make up only 5 to 10 percent of the sources AI search references, so the cited third-party pages belong in the baseline too.
4. Check what AI already sends you. In GA4, filter the referral report for assistant domains such as chatgpt.com and gemini.google.com.

The full routine, prompt sets included, is in how to measure your AI visibility. OnlyAEO is an AEO and GEO agency, and its free AI visibility audit, delivered within 48 hours, covers ChatGPT, Claude, Gemini and DeepSeek.

The cited pages also show where capture rate can grow: our earn-citation list ranks the third-party sites AI cites in your space where you are not mentioned yet, with a recommended play for each.

How we tested this question on ChatGPT and Claude

We start every article by asking ChatGPT and Claude the buyer's question, reading every page they cite in full, and turning each search they ran into a section. For this guide, we asked both assistants the question in its title on 14 October 2026.

• ChatGPT answered from its own knowledge and ran no web search.
• Claude searched for AEO ROI benchmarks and AI search conversion rates, then cited a handful of agency and tool-vendor blog posts.

Neither answer named OnlyAEO, so our own capture rate for this question started at zero. None of the posts Claude cited took gross margin out of its revenue estimate, which is why the formula above includes it.

Should the forecast assume AI-referred visitors convert better?

No. Put your existing site conversion rate in the base case, and use the published evidence as the reason finance should not discount conversion to zero.

The best-known public evidence comes from Adobe, and it covers retail. Adobe's analysis of more than 1 trillion visits to US retail sites found generative AI visitors browsed 12 percent more pages per visit with a 23 percent lower bounce rate.

That report also found AI traffic was 9 percent less likely to convert, improved from 43 percent in July 2024. Adobe's follow-up analysis on its business blog found that in July 2025 an AI-driven visit was worth 27% less than a non-AI visit, versus 97% less in July 2024.

By July 2026, Adobe's August report found AI retail visitors converting at a 60% higher rate than non-AI traffic.

B2B funnels differ from retail checkouts, so present these figures as the upside case and swap in your measured rate once you have one. What an AI-sourced lead is actually worth covers the next step.

How do you run the forecast with your own numbers?

Run the chain once per question cluster for a low, a base and a high case, using the placeholder example below as a template.

EXAMPLE INPUTS: replace every value with your own. A method, not a forecast.
One question cluster, annual figures, any currency.

Input                     Low      Base     High
Annual question volume    40,000   40,000   40,000  (keyword floor)
AI research share         0.15     0.30     0.45
Capture rate              0.05     0.15     0.25
Act rate                  0.10     0.20     0.30
Conversion rate           0.03     0.03     0.03    (your site rate)
Average deal value        12,000   12,000   12,000
Gross margin              0.80     0.80     0.80
Confidence discount       0.60     0.50     0.40

Base case
40,000  x 0.30       = 12,000 questions asked in AI assistants
12,000  x 0.15       = 1,800 answers that name you
1,800   x 0.20       = 360 buyers who click or search for you
360     x 0.03       = 10.8 customers
10.8    x 12,000     = 129,600 revenue
129,600 x 0.80       = 103,680 gross profit
103,680 x (1 - 0.50) = 51,840 discounted expected value

Low case:  3,456     High case: 233,280

Payback: ROI = (discounted expected value - C) / C,
where C is your full annual program cost.

Then run the sensitivity test: move one input at a time to its low value, keep the rest at base, and find the single input that breaks payback.

In the placeholder run, the low case collapses because capture rate and AI research share fall together. If your program pays back only when every input sits at its high value, do not ask for the budget yet.

What timeline should the forecast show?

Show a period of cost with no return first, then leading indicators, then revenue. OnlyAEO's guidance is that the first measurable results usually need 60 days or longer, after which visibility keeps compounding over the months that follow as more pages, videos and mentions get picked up.

Leave out a month-by-month return curve. Finance reads a curve on a slide as a commitment, and the last section explains why you should not make one.

The evidence arrives in stages. HubSpot's review found visibility shifts before traffic does, with earlier gains in AI citations, brand mentions and assisted conversions. Report capture rate against its target first.

If your deals take longer to close than the budget period, modeling AEO payback when your sales cycle outlasts the retainer keeps the forecast from reading as a loss at renewal.

Which comparables help finance trust the number?

Set the expected value next to numbers finance uses elsewhere:

• The equivalent paid click. Look up what the same buyer question costs per click in your paid search account, and express the citation's value as media spend you would otherwise buy.
• Branded search you pay for. Finance already funds the branded clicks where zero-click buyers resurface, so AEO sits upstream of demand the company pays to harvest.
• Pipeline coverage. Express the discounted value as a share of next year's pipeline target.

How do you pressure-test the forecast before finance sees it?

Run the checks a finance reviewer will run:

1. Cost completeness. Include content, technical work, tooling, internal time and any agency fees, because a reviewer will add back whatever you leave out.
2. Attribution honesty. AI assistants often pass no clean referrer, so say now how you will measure value later; proving AEO pipeline without a referrer sets out the methods.
3. Labelled inputs. Next to every input, write its source and name the one your sensitivity test flagged.

What should you promise finance, and what should you refuse?

Promise a method and a schedule, and refuse anything that sounds like a guaranteed result. OnlyAEO gives no results guarantee of its own and promises no particular ranking, citation or placement in an AI answer; an in-house team should hold the same line.

Refuse these:

• A named citation by a date. Answers change from run to run, so no one can promise when an engine will start naming you.
• A fixed conversion rate. Promise the method that will measure the real one.
• Full attribution. Part of the zero-click lift will always go uncounted.

Promise these instead, in this order:

1. A measured baseline of capture rate for the questions in the model, before money is committed.
2. A discounted estimate built on inputs finance helped set.
3. A replacement schedule naming when each estimate becomes a measured number.

For the other side of the table, see how a CFO should evaluate an AEO investment request.

More questions about forecasting AEO ROI

How often does Google show an AI summary at all?

In Pew Research Center's March 2025 browsing study, 18% of all Google searches generated an AI summary. Check which questions in your own set trigger one before you set AI research share for Google traffic.

What does waiting cost in organic traffic?

McKinsey estimated that unprepared brands may see a 20 to 50 percent decline in traffic from traditional search channels. Bain found 80% of consumers rely on AI-written results for at least 40% of their searches.

How widely do consumers use ChatGPT?

Pew Research Center found in 2025 that 34% of US adults had ever used ChatGPT, including 58% of adults under 30. If your buyers skew younger, set AI research share nearer the top of your survey range.

Can measured visibility data feed my own forecast model?

Yes. OnlyAEO data is available through its MCP and API for Claude, Cursor, n8n and other tools, so you can pull it into your own model.