Ask HerculesAsk Hercules← All posts

    November 21, 2026

    How agencies can prove the ROI of AI search optimization

    Show clients what AI search optimization delivers. Connect citations to qualified inquiries, track costs and separate observed results from modeled ROI.

    How agencies can prove the ROI of AI search optimization

    By Colleen Nagle, Founder of Muse Platforms and Ask Hercules

    Agencies can demonstrate the value of AI search optimization by connecting documented content changes to measured visibility, qualified inquiries and customer outcomes. To claim financial ROI, they also need costs, attributable economic benefit and a defensible comparison. Citation growth is useful evidence of visibility; it does not, by itself, prove revenue or return on investment.

    A client asks a fair question: “We are appearing in more AI answers. Is that helping the business?”

    The answer should survive a follow-up from the person who approves the budget. Show what changed, what the evidence supports and what remains unknown. If revenue attribution is incomplete, report the visibility and pipeline evidence without filling the gap with an impressive-looking number.

    At Ask Hercules, our recommendation is to build the measurement plan before the content plan. That makes it easier to choose work that can be evaluated and to explain its commercial purpose.

    Agree on the buying problem first

    Companies hire AEO services when they need help understanding where they appear in AI-assisted discovery, correcting inaccurate descriptions or improving the information available to prospective buyers. Some also lack the people or time to implement and test the necessary changes.

    Define the business objective more narrowly than “increase visibility.” For a consultancy, it might be qualified conversations about a specific engagement. For a SaaS company, it might be demos from organizations that fit its target customer profile.

    Agree on what qualifies. A useful inquiry should have a relevant need and a plausible fit with the offer. A bot submission or a request for an unrelated service should not count as a commercial win.

    Build a baseline that can be repeated

    Choose a fixed set of relevant buyer questions. Record question wording, geography, platform, model or test surface, retrieval mode and dates. Repeat the baseline to see how much results vary before making changes.

    Retain the answers and the source URLs. Identify provider failures separately from valid answers that omit the brand. An outage is a data-quality problem, not evidence that a business lost visibility.

    When adding new questions, report them as a separate cohort or rebuild the baseline transparently. Comparing fifty easy branded questions this month with fifty difficult category questions last month can produce a dramatic chart with little useful meaning.

    Report four kinds of evidence separately

    Measurement layer Useful evidence What it can support
    Website readiness Access issues resolved and approved page changes The site is better prepared within the checks performed
    Observed AI visibility Saved answers, accurate mentions and relevant citations The brand or page appeared in the sampled responses
    Buyer action Relevant visits, qualified inquiries and booked meetings People took commercially meaningful next steps
    Economic outcome Closed business, contribution margin and total program cost A financial evaluation under stated attribution assumptions

    Use definitions consistently. A citation to a third-party article about the client is different from a citation to the client's own domain. Both may matter, but they should not silently become the same metric.

    Also review accuracy. An assistant that names the brand but invents a service is not a clean success. Include incorrect descriptions in the action list.

    Explain the denominator behind every percentage

    Suppose a generic test receives 100 valid responses and 20 contain a verified link to the client's domain. The pooled citation rate is 20%. If 10 additional calls failed, record those failures separately; do not treat them as ten answers without citations.

    Label the aggregation method. Ask Hercules' documented Citation Score averages per-question scores across answering models, which can differ from a pooled response rate when provider availability varies. Use the product's definition when reporting its score. [1]

    Show the numerator, denominator and test context beside the result. A score becomes more useful when the client can trace it to the underlying answers.

    Connect discovery to the sales process

    Use web analytics to capture identifiable referring sources and meaningful conversions. In the CRM, record the inquiry, qualification outcome, opportunity and final sale. Preserve first-touch and later-touch information where available rather than repeatedly overwriting a single source field.

    Add a simple question to the intake process: “Where did you first hear about us?” If a prospect names an AI assistant, save that as self-reported evidence. It can help explain a journey that otherwise looks like direct traffic or a later branded search.

    Google states that traffic from its AI features is included in Search Console's Web performance reporting. A change in total Google traffic therefore does not isolate the effect of an AI Overview. [2]

    Report identified AI referrals, self-reported AI discovery and unattributed journeys separately. Deduplicate the people and opportunities appearing in more than one category before totaling outcomes.

    Use a financial example that exposes its assumptions

    The following numbers are hypothetical, not Ask Hercules or client results. Suppose a three-month program costs $6,000, including agency fees, software and internal implementation time. Six closed customers are attributed to the program, each contributing $2,000 after the direct costs of delivering the sale. Total contribution is $12,000.

    Under those assumptions:

    ROI = (attributed contribution – program cost) ÷ program cost

    ROI = ($12,000 – $6,000) ÷ $6,000 = 100%

    Now test the fragile assumption: attribution. If only three customers can reasonably be credited to the program, contribution falls to $6,000 and ROI is 0%. If no incremental customer effect can be established, the calculation is an attribution scenario, not proven incremental return.

    Use a consistent accounting period. For subscriptions, state whether contribution is realized during the period or projected over a defined horizon. Do not substitute uncollected lifetime revenue for realized contribution without clearly labeling the forecast and its assumptions.

    Pipeline has value as an early indicator, but an open opportunity is not a sale. Report pipeline separately from realized ROI.

    Strengthen the case for cause and effect

    Keep a change log identifying the page, the edit, the publication date and the expected effect. Where practical, stagger changes across comparable pages or question groups and monitor a comparison group that has not yet received the work.

    That approach can strengthen an evaluation, but it does not eliminate every confounding factor. Product launches, paid campaigns, seasonality and changes in AI systems may affect the same period. Disclose those factors rather than attributing every movement to AEO.

    For low-volume B2B businesses, review a longer period and show the actual counts. Moving from one inquiry to two is a 100% increase; it is also one additional inquiry. Both descriptions belong in the conversation.

    Use Ask Hercules for the visibility evidence

    Ask Hercules provides an Audit, Fix and Track workflow with question-level response evidence, citation and mention distinctions, and repeated measurement. An agency can use that evidence to connect an observed gap with a specific remediation and examine later responses. [1]

    Use your analytics and CRM alongside it for customer and financial outcomes. Do not present a visibility dashboard as an automatic revenue attribution system.

    Muse Platforms' Subamerica case study illustrates why interpretation matters. The supplied August and September reports use different GEO question sets, so their visibility percentages are not a clean like-for-like trend. They also do not prove lead or revenue lift. Those limitations make the reporting more useful because they show exactly where the evidence ends. [3]

    Give the client a decision they can make

    Close the report with the next action: continue a useful change, improve a weak landing page, investigate inaccurate answers or stop work that has not justified its cost.

    A practical monthly narrative answers four questions in plain language: what we changed, what happened, what the evidence can establish and what we recommend doing next. Include the supporting data for anyone who wants to examine it.

    Frequently asked questions

    Can an agency promise a specific number of AI citations?

    It can define the work, test conditions and reporting cadence. It should not present external AI responses as a guaranteed deliverable it controls.

    What if citations improve but leads do not?

    Check question relevance, the cited page, offer clarity and conversion friction. Also consider the sales cycle and sample size. Better visibility may not yet have produced a measurable commercial effect.

    How should we present share of voice?

    State the competitor set, question set and formula. A share measured within a selected test panel is not the brand's share of every AI answer in the market.

    Build a report your client can question

    Explore the Ask Hercules methodology for the evidence behind its visibility metrics. For help connecting SEO, AEO and GEO delivery with a practical measurement plan, talk to Muse Platforms. Start by defining the customer action that would make the work worthwhile.

    Sources

    [1] Ask Hercules — Methodology.

    [2] Google Search Central — Measuring performance in AI search features.

    [3] Muse Platforms — Subamerica case study.


    See what AI says about you

    Ask Hercules is a product of Muse Platforms, Inc.

    Want this analysis for your site?

    Book a free AI Visibility Review with the Ask Hercules team.

    Book your free AI Visibility review