Quick answer: You can't measure AEO with click attribution, because the value it creates – your brand named inside an AI answer – usually never produces a click to attribute. Measure it in layers instead: citation presence and share of voice across your priority queries, visible within the first 90 days; corroborating demand signals like branded search and self-reported attribution, which build over two to four months; and pipeline lift, which shows up on a three-to-six-month horizon. Because none of that value arrives as a click, pricing the work purely on performance under-counts it – which is why more of these engagements now run on flat or hybrid fees.
Why can't click attribution measure AEO?
Click attribution was built to reward a link somebody follows, and a growing share of AEO's value never produces one to follow. When an AI engine answers a comparison or "best X" question directly and names your brand as part of that answer, the buyer's job is frequently done at that point. There's no next step that generates a session in Google Analytics, no referral row in your dashboard, no line item anywhere for a decision that got made without a click.
In practice, this shows up as an ambiguous pattern rather than an obvious one: a term holding a strong position with real impression volume and almost no clicks behind it. That pattern has more than one honest explanation – a snippet that isn't earning the click, ordinary ranking noise, or the answer genuinely being resolved upstream by an AI engine – and your analytics platform can't distinguish between them on its own. The only way to know which explanation applies is to go check the engines directly on those terms: query them, and record whether your brand is named. That audit is the subject of the next two sections.
There's also a version of this problem that isn't about ambiguity at all – it's about whether a citation exists to check in the first place. Similarweb clickstream data showed the share of ChatGPT answers carrying a citation rose from 0.6% in January 2025 to 2.8% by August 2025 (the underlying report itself published in 2026). Citation rates are climbing fast on a relative basis, but the absolute share remains small. That matters for measurement because it means the well-known "a citation doesn't always produce a click" problem is really the smaller of two gaps: most of the time, an AI answer doesn't generate a visible citation link at all, whether or not your content shaped the answer behind the scenes. A measurement plan built only to catch clicked citations is missing almost everything.
This shows up across every vertical we manage programs in – insurance, banking, commerce, home services – and the underlying mechanism doesn't change by category. What changes is how often a category's buying questions get resolved this way, which is exactly why building a query set matched to your own category, rather than borrowing someone else's, is the first real piece of work below.
What can you track in the first 90 days?
Three things are trackable inside a single quarter, and none of them require waiting on a sales cycle to close.
First, citation presence. Run your priority queries through AI Overviews, ChatGPT, Perplexity, and Copilot on a weekly cadence and record, query by query, whether your brand is named or cited. This is binary and simple to log: yes, no, and which engine.
Second, share of voice. Once you're tracking presence across a query set, roll it into a percentage – what share of category answers name you versus your named competitors. This is the number that tells you whether you're winning the category conversation, not just occasionally showing up in it.
Third, citation source mix. When you are cited, what's actually doing the citing – your own site, a partner site, a review property, an editorial piece? This is where the audit connects back to your broader content and partner footprint, and it's worth checking against the corroboration bias the AEO pillar describes: these engines favor a source that isn't self-interested, so if your presence traces mostly to partner and editorial content rather than your own domain, that's the expected shape, not a problem to fix. Confirming that for your own footprint doesn't require proprietary data or special tooling – it takes running the audit once and tagging what shows up.
How do you build the query set that makes this work?
None of the above functions without a query list, and building one well is an afternoon of work, not a quarter-long project.
Start with two segments you likely already have visibility into through your own search reporting. The first is striking-distance terms – queries where you're ranking outside the top results but still in range, since these are terms where an AI engine is plausibly already weighing your content without yet citing it. The second is high-impression, low-engagement terms – queries with real search volume attached but little traffic to show for it. Both segments are worth including precisely because they're ambiguous on their own: some genuinely reflect an answer being resolved upstream, some are underperforming for ordinary reasons, and the citation check – not your search reporting alone – is how you find out which is which.
Second, add category-intent questions that don't come from your existing keyword data at all: "best [category] for X," "[you] vs [competitor]," "how do I choose a [category] provider." These are the exact question formats AI engines are built to answer directly, and they're often absent from a paid-search or SEO list because they were never worth bidding on or ranking for under the old model.
Third, add the questions your sales team already hears. Every AE and every intake call surfaces phrasings prospects use while comparing options: "why should I use you instead of," "is [category] worth it," "what's actually the difference between." That language sits with sales and rarely makes it to marketing on its own. Ask for it directly, in a real conversation, not a survey.
On sizing: a query set in the 30-to-60 range is the workable middle ground for most single-category brands. Below 15 queries, ordinary single-query noise – an engine flips its answer for one term one week – reads as a program shift when it isn't. Above 100, manual tracking quietly stops happening every week, which defeats the point of a leading indicator. Build the first version around the segment of your category that matters most commercially, then expand once the process is running.
Refresh the list on a cadence, not whenever someone remembers to. Quarterly is the right default, since category questions shift as competitors launch, as your product line changes, and as the engines themselves change what counts as a "best X" answer worth generating. Trigger an off-cycle refresh any time a new competitor enters your top five, or any time sales flags a new comparison question showing up in calls.
Keep the tracking sheet itself simple: one tab, with columns for the query, its source (striking-distance, low-engagement, category question, sales-sourced), and the engine it's checked against. That tagging is what tells you, at quarter-end, whether gains are concentrated in terms you already had ground on or in the terms you added deliberately – the difference between defending existing territory and actually expanding into new ground.
Which demand signals corroborate citation activity?
Citation metrics tell you the input is happening. The next layer, checked monthly rather than weekly, tells you whether it's translating into demand.
Branded search lift is the cleanest of the three. Pull branded query volume and watch it against your citation-presence trend. If share of voice is climbing while branded search stays flat, something's off – either the citations aren't reaching the audience that matters, or the query set doesn't match where your real buyers actually search.
Self-reported attribution is the second signal, and most teams under-use it. Add "how did you hear about us" to intake forms and sales qualification, with AI tools listed as an explicit option alongside search, referral, and social. Prospects increasingly answer with some version of "I asked ChatGPT" or "it came up while I was researching," which is direct corroboration no analytics platform hands you on its own.
Third is a quality shift in existing inbound metrics you're probably already tracking. WiserAdvisor's program, for instance, already reports lead-to-engaged-lead conversion at 44%. That's not a number AEO produced – it's a metric that predates any citation work – but watched over time alongside a rising share-of-voice trend, a lift in that kind of downstream quality metric is exactly the corroborating signal this tier is built to catch: buyers arriving further along, already informed by something other than a first search result.
When does AEO show up in revenue?
Months three through six is the typical window, and three separate mechanics stack up to produce that lag.
The first is engine-side: AI systems re-crawl and re-weight sources on their own schedule, not yours, and you don't control when a citation you've earned gets reflected in the next answer generation.
The second is compounding corroboration. Branded search lift, self-reported attribution, and inbound quality shift each build gradually and reinforce one another; none of them moves in a single clean step the way a paid click does.
The third is the ordinary buying-cycle lag present in any considered purchase, regardless of channel. A buyer who sees your brand named in month one may not be in-market until month four.
None of this means last-click attribution stops working. It still captures the bottom of the funnel the way it always has, and that engine doesn't need re-explaining or replacing. What it can't see is the discovery layer sitting above it – and Tier 3 measurement usually doesn't require new instrumentation to start watching for it, just a link back to which content and query clusters preceded a metric you already track. Unlock's home-equity program, for example, already reports an account-creation-to-application rate of roughly 20% every month. That number exists whether or not anyone is running a citation audit. The measurement discipline here is tying that existing number back to which clusters and which content were active in the citation checks that preceded it that quarter – not building a new metric from nothing.
Why run three tiers instead of one?
Stack the metrics above into three tiers on three different clocks, because each tier answers a different question and each one, watched alone, produces a specific and predictable failure.
Tier 1 is citation metrics – presence, share of voice, source mix – updated weekly. It's the leading indicator: whether the program is showing up in the answers that matter, well before there's any way to know whether that's translating into pipeline.
Tier 2 is the corroborating demand signals – branded search lift, self-reported attribution, inbound quality shift – updated monthly. It won't move week to week, and shouldn't be checked that often, but a month is enough time for a real trend to separate from noise.
Tier 3 is business outcomes – pipeline and revenue lift, tied back to the specific content and query clusters driving citations – updated quarterly. It's the tier that answers the actual question a CEO asks, and also the slowest one and the easiest to misread sitting on its own.
Watch only Tier 3, and a working program gets cut at the ninety-day mark because revenue hasn't moved – the same mistake pure click attribution invites, since rising Tier 1 presence is invisible to anyone checking quarterly revenue alone. Watch only Tier 1, and a program keeps getting funded for showing up in answers that never turn into demand, mistaking visibility for value. Watch Tier 2 or Tier 3 without Tier 1 as context, and growth that actually came from a seasonal lift, a new sales rep, or a competitor stumbling gets credited to AEO, with no leading indicator available to check the timing against. Running all three together, on their own cadences, is what lets you tell a program that isn't working yet from one that isn't working at all.
How do you pay for value a click can't price?
Performance pricing assumes a click exists to measure and pay against. Citation value structurally doesn't produce one, which means a program priced purely on clicks or conversions will always undercount what AEO is contributing – by design, not by accident.
That's why flat and hybrid fee structures now show up in most proposals for this work: a flat retainer, or a retainer against a percentage of program performance, whichever is greater in a given period. The flat component ties directly to the Tier 1 metrics above – presence across priority queries, the share-of-voice trend, and whether citation-worthy content is kept fresh enough to hold those citations as engines re-crawl. That's a different accountability structure than "did this produce a click," but it's a real one, and it's measurable using exactly the tracking described in this piece.
The fuller argument for why an affiliate program specifically is the asset worth funding this way – since the same content that earns affiliate placement is often built to the sourcing and structure standards these engines pull from – is the pillar piece this measurement plan sits underneath. For the pricing mechanics behind a flat or hybrid structure, see our breakdown of what program management actually costs and what drives the number.
What does this actually cost to run, and who owns it?
The honest answer is smaller than most CMOs expect. The weekly cadence – running your query set through AI Overviews, ChatGPT, Perplexity, and Copilot, logging presence and citation source – is manual work, and a spreadsheet is genuinely fine to start with. For a query set in the 30-to-60 range, budget two to three hours a week for someone to run the queries, log results, and roll them into a share-of-voice trend. That's a task, not a role, at that scale, and it fits inside an existing analyst's or SEO manager's week without a new hire.
Tooling earns its cost once you cross a real threshold: your query set grows past 75 to 100 terms, you're tracking consistently across more than two or three engines, or you need historical trend data further back than a spreadsheet comfortably holds. Below that threshold, tooling is usually solving a problem you don't have yet. Above it, the manual version quietly stops happening every week, which defeats the purpose of a leading indicator.
On ownership: in-house, this sits naturally with a marketing analyst or SEO function, since weekly Tier 1 tracking is close-cousin work to rank tracking they may already do, and the monthly Tier 2 rollup slots into an existing reporting cadence. Run through an agency, it typically folds into program management rather than billing as a separate line – the same team managing affiliate or paid channels absorbs the weekly and monthly work, with the quarterly Tier 3 review as a standing agenda item rather than a bespoke project. The one arrangement to avoid is splitting the three tiers across three people who never compare notes; the point of the structure is that someone is looking across all three cadences, not just running their own tier in isolation.
Frequently asked questions
What KPIs actually measure AEO, and on what cadence? Citation presence, share of voice, and citation source mix, tracked weekly, are the leading indicators. Branded search lift, self-reported attribution, and inbound quality shift, tracked monthly, are the corroborating layer. Pipeline and revenue lift tied back to citation-driving content, tracked quarterly, is the outcome layer. No single KPI on its own answers the question of whether the program is working.
How many queries should be in the tracking set, and where do they come from? Thirty to 60 queries is the workable range for most single-category brands. Seed the list from striking-distance search terms and high-impression, low-engagement terms, then add category "best X" and "vs" questions plus the comparison language your sales team already hears. Fewer than 15 queries and single-query noise reads as a trend that isn't real. More than 100 and manual tracking breaks down unless tooling is already in place.
How do you tell "not working yet" from "not working"? By checking all three tiers on their own clocks rather than any single one in isolation. Citation presence moving in the first 90 days without any Tier 2 movement by month four is a real warning sign. Tier 2 movement with flat Tier 1 presence suggests something other than AEO is driving it. Neither tier moving at all after a full quarter is the honest signal that the program needs a different approach, not more patience.
Do you need special software, or is a spreadsheet enough to start? A spreadsheet is enough for a query set in the 30-to-60 range tracked across three or four engines. Add tooling once the query set outgrows that range, engine coverage needs to expand past what's manually sustainable, or you need historical trend data further back than a spreadsheet reasonably holds.
Who on the team should own this, and does it change if an agency runs the program? In-house, it belongs with whoever already owns SEO or marketing analytics for the weekly tier, and whoever owns branded search and lead-source reporting for the monthly tier; the quarterly review should include whoever owns pipeline reporting, since that's the tier that closes the loop back to revenue. If an agency runs the program, this typically folds into existing program management rather than requiring a new internal hire.
How is measuring AEO different from tracking SEO rankings? A rank tracker checks where a link sits on a results page most users no longer have to visit to get an answer. A citation check asks a different question entirely – was your brand named inside the answer itself – and it can only be answered by directly querying the engines, not by inferring it from ranking position or from click volume. The two disciplines run in parallel, on different clocks, and neither substitutes for the other.
If your affiliate program or your own content is already earning citations, the real question isn't whether AEO is happening. It's whether you've built a way to see it, on a timeline honest enough to judge it by, before you decide whether it's working.