Measurement

What does an AI citation actually cost, compared to ads?

By Arnav Mukherjee, founder of TofuBofu · July 31, 2026

TL;DR

A business services click now costs a median $5.87 and a lead $93.69, across 13,474 campaigns. That cost recurs forever. A citation is paid once and then repeats free until it decays, which makes per-unit comparison the wrong frame. The real problem is concentration: in our data, engines named 1,329 companies but the top 10 took 42% of all mentions and 64% of companies appeared exactly once. Appearing is cheap, staying is expensive. And no honest ROI number exists yet, because nobody runs holdouts and AI referral tracking structurally undercounts. Here is the break-even maths to run on your own numbers instead.

Every founder who has looked at AEO seriously asks the same question within about five minutes, and it is the right question: what does this cost me per customer compared to just buying the clicks? I get asked it on sales calls constantly, usually by people who already run paid search and know their cost per lead to the dollar.

The answer nobody in this category wants to give is that the comparison does not work the way you want it to, and that the honest version is more useful than the confident one. I am going to give you the real ad benchmarks, the actual cost structure of a citation, our own data on why the distribution is brutal, and then the part where I tell you that the ROI number you want does not exist and why you should distrust anyone who hands you one.

What a click costs in 2026

Start with the number you can actually trust, because it comes from a large sample with a published methodology. WordStream's 2026 benchmarks analysed 13,474 US search advertising campaigns between April 2025 and March 2026, reporting medians rather than means so outliers do not distort the figure, with a minimum of 52 active campaigns per subcategory.

Segment	Median cost per click	Median cost per lead
Business services	$5.87	$93.69
All industries	$5.42	$66.69

WordStream 2026 Google Ads Benchmarks. 13,474 US search campaigns, April 2025 to March 2026, reported as medians.

Those numbers have two properties worth naming before we compare anything. They are predictable, which is the entire product: you can decide on Monday to spend money and have leads by Friday, with a cost that is knowable in advance. And they are permanently recurring. There is no accumulation. The two hundredth lead costs the same as the first, and the day you stop paying, the flow stops completely. You have rented attention, not built anything.

What a citation costs, honestly

Here is where most comparisons cheat by dividing content spend by citations and printing a tidy "cost per citation." I will not, because the denominator is unstable in a way that makes the resulting number meaningless. Instead, the actual structure.

A citation costs you the production of an asset that deserves to be cited, plus the corroboration that makes an engine trust it. The asset is a knowable cost: a properly researched comparison page or a genuinely useful answer to a buying question, whether you write it, hire it out or generate and edit it. The corroboration is the harder half, because it cannot be bought directly. Review-site presence, community mentions, being referenced by third parties. Research from Profound found brands present on four or more platforms are 2.8x more likely to be cited, and SE Ranking found 71% of ChatGPT-cited pages use structured data. Those are the inputs, and neither has a price list.

Then the shape diverges completely from ads. Once the asset earns citations, each additional citation is free. The engine names you a hundred times and you pay nothing more. The asset does decay, as models update and competitors publish better answers, so it is not an annuity, but it decays over months and quarters rather than stopping the instant you stop paying. You are buying a depreciating asset instead of renting a meter.

The part that makes it expensive: it is winner-take-most

This is the finding that changed how I think about the economics, and it comes out of our own scan data. Across 75 completed scans covering 44 domains, we recorded every company that engines named in response to buying questions. That produced 7,333 individual company mentions spread across 1,329 distinct firms.

The distribution is not close to even. The top 10 companies accounted for 42% of all mentions. Meanwhile 851 companies, which is 64% of every firm the engines named, appeared exactly once across the entire dataset.

That is a steep power law, and it has a direct cost consequence. Getting named once is cheap and achievable: two thirds of the companies in our data managed it, presumably by accident. Getting into the small group of names that appear consistently, which is the only version that produces reliable pipeline, is expensive, because it needs the third-party corroboration layer that money cannot buy directly. So "what does a citation cost" splits into two very different questions with two very different answers, and every vendor pitch quietly answers the cheap one.

The same data gives the sobering baseline: the median company we scanned was named on 8% of its buying-intent question-and-engine combinations, and 31% of scans came back at exactly zero. That sample is self-selected, because people who run a visibility scan usually suspect they have a problem, so read it as a floor rather than an industry average. But it is a useful corrective to any pitch that implies citations arrive shortly after you publish something.

Find out which side of the distribution you are on

A free scan shows how often six AI engines name you on real buying questions, and which competitors take the mentions you are missing.

Run your free scan

Why nobody can give you a real ROI number

You will find blog posts quoting a precise ROI multiple for AEO against ads, and conversion benchmarks for AI-sourced leads stated to one decimal place. I went looking for a defensible one to put in this post and could not find it. Three structural reasons, and they are not going away soon.

Nobody is running holdouts. Not us, not the enterprise platforms. Without a group of comparable companies deliberately left untreated, every "we improved visibility and revenue rose" claim is an uncontrolled before-and-after, made in a period when the company was almost certainly also changing its content, its pricing and its outbound motion. Attributing the revenue move to the citation count is a choice, not a finding.

The denominator leaks badly. AI referral traffic structurally undercounts. Zero-click answers resolve the buyer's question without a visit at all. Many AI surfaces do not pass a clean referrer, so visits that do happen land in direct traffic. And the very common path, where a buyer reads an AI answer and then searches your brand name, arrives as branded search. Every AI traffic number is a floor, not a total. You cannot compute a trustworthy cost per lead on a denominator you cannot see.

The identifiable slice is unrepresentative. This is the subtle one and it is why I distrust published AI-lead conversion rates specifically. The AI-sourced visits you can identify are the ones that happened to pass a referrer, which is a biased subset of all AI-influenced buyers. Measuring conversion on that slice and presenting it as the conversion rate of AI-sourced leads is a sampling error dressed as a benchmark.

There is a plausible mechanism for AI leads converting well, and I believe it directionally: a buyer arriving from an AI recommendation has been pre-filtered by the engine and typically arrives later in their research, which has always correlated with better conversion. G2's 2026 research found 69% of buyers switched a vendor decision based on AI input, and one in three ended up with a vendor they had not heard of before AI recommended it, which tells you the influence is real. But "the influence is real" and "the conversion rate is 4.2%" are very different claims, and only the first is currently supportable.

The break-even maths to run instead

Since the industry number does not exist, use your own. This will not forecast your return, but it sizes the bet in units you already understand and usually settles the decision in about ten minutes.

1. Start from your real cost per lead, not the benchmark

Use what you actually pay, not the $93.69 median. If you pay $140 per lead on paid search, that is your reference price. The benchmark is only there to tell you whether your own number is unusual.

2. Convert the AEO investment into lead-equivalents

Take your planned quarterly AEO spend, tools plus content production plus your own time costed honestly, and divide by your cost per lead. If the quarter costs $6,000 and your leads cost $140, the investment has to beat roughly 43 leads. That is the hurdle, stated in a currency you already trust.

3. Ask how many questions you would need to win

Work backwards. Given your deal size and close rate, how many buying-intent conversations does 43 leads represent, and how many of your buying questions would you need to be named on to plausibly produce them? If the answer is 'nearly all of them', the maths is not close and ads are the better use of that money this quarter.

4. Price in the decay, not an annuity

Do not model citations as permanent. Assume the asset degrades over several quarters as models update and competitors publish. The honest question is whether it outlives the period during which the same money would have been spent on clicks, which for most content is a genuine maybe rather than an obvious yes.

5. Separate 'appear once' from 'appear consistently'

Budget for the second one or do not start. Our data shows 64% of named companies appeared exactly once, which is what an underfunded effort buys: a scattering of mentions that never becomes a position. The top 10 firms taking 42% of mentions is the target, and it is a materially larger commitment.

6. Decide what you are actually buying

Ads buy attention while you have money. Citations buy an asset that keeps working after the money stops, at the price of a slower and less certain start. If you need pipeline this month, that is not an AEO problem and AEO will not solve it in time.

The two channels are not really substitutes and framing them as competitors leads to bad decisions in both directions. Ads are the only channel that delivers on a schedule, and a business that needs revenue this quarter should buy clicks without guilt. AEO is the thing you do so that in four quarters you are not paying $5.87 every single time someone wants to find a company like yours. Most firms that can afford both should run both, and the ones that can only afford one should pick based on how long they can wait, not on which has the better story.

What I would not do is pick based on a published ROI multiple, because the honest state of this category is that the intermediate outcome is precisely measurable and the revenue outcome is not yet. You can know, today, whether AI names you when a buyer asks. That is a real fact, it is free to check, and it is a more solid basis for a decision than a number somebody computed without a control group.

Frequently asked questions

What does an AI citation cost compared to a Google Ads click?

They have different cost shapes, which is why a direct per-unit comparison misleads. WordStream's 2026 benchmarks, drawn from 13,474 US search campaigns, put the median business services cost per click at 5.87 dollars and cost per lead at 93.69 dollars, and that cost recurs for every single click forever. A citation is paid for once, in the production cost of the asset that earns it, and then keeps appearing at no marginal cost until it decays. The honest framing is not price per unit but whether your asset gets cited enough times before it decays to beat what the same money would have bought in clicks.

Is AEO cheaper than paid ads?

Cheaper per outcome if it works, more expensive if it does not, and far less predictable either way. Ads are a purchase: you pay a known price and get a known volume, starting today and stopping the day you stop paying. Citations are an investment with an uncertain payoff and a long lead time, and in our own data most companies are not cited at all on buying questions, with 31 percent of scans returning zero. The correct comparison is not cheaper versus dearer, it is a predictable recurring cost against an unpredictable asset that compounds if it lands.

What is the ROI of AEO compared to ads and email?

Nobody can currently give you a trustworthy number, and you should treat anyone who does with suspicion. There are three reasons. No vendor in this category runs holdout groups, so no causal claim has a control. AI referral traffic structurally undercounts because of zero-click answers, missing referrers and branded-search follow-ups, so the denominator is unknown. And the companies investing in AEO are usually changing other things at the same time. You can measure the intermediate outcome, whether AI names you on buying questions, precisely. You cannot yet measure the revenue effect cleanly, and honest vendors say so.

Why is getting cited winner-take-most?

Because AI recommendations are a short list, not a page of ten blue links. In our first-party data, engines named 1,329 distinct companies across 7,333 total mentions, but the top 10 companies alone accounted for 42 percent of every mention, and 64 percent of named companies appeared exactly once. That is a steep power law. The practical consequence for cost is that appearing once is cheap and relatively easy, while entering the small set of names that appear consistently is much more expensive, because it requires third-party corroboration you cannot buy directly.

How do I calculate the break-even on AEO investment?

Work backwards from your own cost per lead rather than from any industry average. Take what you currently pay per lead on paid search. Divide your planned AEO investment for a quarter by that number to get the lead-equivalent the investment has to beat. Then ask how many buying-intent questions you would need to be named on to plausibly produce that many conversations, given your deal size and close rate. This does not give you a forecast, because the payoff is uncertain, but it tells you the size of the bet in units you already understand and it exposes immediately whether the maths is even close.

Should a company with no marketing budget do AEO or ads first?

Ads if you need pipeline this month and can afford a recurring cost, because they are the only channel that delivers on a schedule. AEO first if your problem is that you are structurally invisible and your budget cannot sustain a permanent per-click spend. The two are not really substitutes: ads buy attention while you have money, and citations accumulate an asset that keeps working after the money stops. The failure mode worth avoiding is treating AEO as a cheaper ads replacement with the same time horizon, because the time horizons are not comparable.

Do AI-sourced leads convert better than paid search leads?

There is a plausible mechanism, but no reliable benchmark, and we will not invent one. The mechanism is that a buyer arriving from an AI recommendation has been pre-qualified by the engine and often arrives later in their research, which historically correlates with better conversion. Against that, the measurement is badly broken: many AI-influenced visits never carry an AI referrer at all and land in direct or branded search, so the visits you can identify as AI-sourced are an unrepresentative slice. Anyone publishing a precise AI-lead conversion benchmark today is reporting on that unrepresentative slice.

Sources and further reading

WordStream 2026 Google Ads Benchmarks: 13,474 US search campaigns, April 2025 to March 2026, medians. Business services $5.87 CPC and $93.69 CPL; all industries $5.42 and $66.69
TofuBofu first-party scan data: 7,333 company mentions across 1,329 distinct firms, from 75 completed scans covering 44 domains, June 24 to July 27, 2026. Self-selected sample skewed to B2B services and SaaS
G2 2026 AI Search Insight Report: 69% of buyers switched a vendor decision based on AI, and one in three chose a vendor they had not heard of before AI recommended it
Profound research: brands present on four or more platforms are 2.8x more likely to be cited
SE Ranking AI search study: 71% of ChatGPT-cited pages use structured data