You can’t prove AI search ROI. Size the cost of AI invisibility instead.

Here is the uncomfortable truth about AI search ROI: you cannot prove it the way you proved SEO ROI. There is no referral tag for a recommendation. A buyer asks ChatGPT for the best option in your category, reads your name in the answer, and remembers it three days later when they buy. No analytics tool connects that sale back to the AI. The click your attribution model depends on never happened.

I want to be clear this is not a BlueJar limitation we are tiptoeing around. It is the nature of the medium, and most searches already work this way. The same Bain and Dynata research found that about 80% of consumers rely on AI-written results for at least 40% of their searches, and roughly 60% of searches end without a click to any website. You can spend real money being recommended by AI and have nothing in Google Analytics to show for it.

Which is why I think agencies and brands keep selling the wrong number. They pitch “AI will send X% more traffic,” a forecast nobody can verify after the fact. I want to make the case for the opposite move. Stop sizing the invisible upside. Size the measurable downside instead: the cost of AI invisibility, right now, in dollars. That is the number that survives a budget meeting.

Why AI search ROI cannot be proven the old way

Traditional digital marketing ROI rests on the click. A buyer searches, clicks your result, lands on a tagged page, and your analytics stitch the journey together. Rank moved, traffic moved, revenue moved. You could draw the line.

AI answers break that line in two places. The engine usually answers in place, so there is no click to a website. And even when the buyer does act, the influence happened inside a conversation you never see. As we covered in our piece on the zero-click crisis in AI search, the click was the carrier signal for attribution, and the carrier is fading.

The pressure is real, and worth dating so you can judge it for yourself. A February 2024 Gartner forecast predicted traditional search engine volume would fall 25% by 2026 as AI chatbots become substitute answer engines. That is a directional prediction, not a measured fact, but the direction is not in dispute. On click-through, an Ahrefs study measured how Google AI Overviews compress clicks. An earlier cut found a 34.5% drop in click-through rate for the top result. A later analysis of 300,000 keywords put the drop at 58% for the top-ranking result on queries where an AI Overview appears.

Look at what that does to your ROI model. The buyer reads your brand in the answer and never generates the click your spreadsheet was built to count. You are not failing to drive ROI. You are failing to measure it, which is a different problem with a different fix.

I will say the honest part plainly, because it is the spine of this whole argument. BlueJar does not do AI-referral attribution. We do not claim to track the clicks AI sends, because in most cases there are none to track. What we measure is what the AI says about you: whether it names you, whether it cites your page, where it names a competitor instead. That is a measurable thing. The clicks are not.

Size the downside you can measure, not the upside you cannot

If you cannot measure the upside, measure the downside. The downside of AI invisibility is countable in a way the upside never is. Every category has a set of buying queries people ask before they choose: “best X for Y,” “X vs Z,” “who should I use for this job,” “is X worth it.” When a buyer asks one of those and the AI names a competitor and not you, that is a specific, observable loss. You can see it. You can count it.

And buyers act on those answers. The March 2026 G2 research, a survey of 1,076 B2B decision-makers, found that 51% now begin software research with an AI chatbot more often than with Google, 69% chose a different vendor than they had originally planned based on AI chatbot guidance, and one in three purchased from a vendor they had not heard of before the AI surfaced it. That last figure is the one that should change how you think. A ready buyer can pick a company they never considered, purely because the AI named it and not the incumbent.

The same thing is happening in consumer categories. A July 2025 Omnisend survey of 1,224 US consumers found 59% already use generative AI tools for shopping tasks, and one in four say ChatGPT product recommendations beat Google’s. For scale, OpenAI reports ChatGPT passed 800 million weekly active users in late 2025. This is not a fringe channel you can afford to be invisible in.

Here is the reframe in one table. The upside column is what the category keeps trying to sell. The downside column is what you can actually put a number on.

Query	Prove the upside (you cannot)	Size the downside (you can)
What are you measuring?	Extra traffic or sales AI “sends” you	Buying queries where AI names a competitor and not you
Is there a click to track?	Usually no, the answer happens in place	Not needed, you read the answer directly
When do you know?	Never, there is no referral tag for a recommendation	Today, by running the queries across the engines
What is the output?	A forecast nobody can verify	A count of lost queries times customer value
Will finance approve it?	Rarely, it reads as “we might be missing AI traffic”	Yes, it reads as “we are losing $N in known demand”

This is what turns AI search into a business case instead of a hunch. You are not promising traffic. You are pricing a leak you can point at.

How to build the cost-of-invisibility number

The method is simple enough to run by hand for one category, and it is the same logic BlueJar automates at scale. Four steps.

• List the real buying queries. Not generic keywords, the actual phrasings a ready buyer uses: “best [category] for [use case],” “[you] vs [competitor],” “who should I hire for [job] in [city],” “is [product] worth it.” Pull them from how customers actually talk, not from a keyword tool’s auto-suggestions.
• Run each query across the engines. Ask the same queries in ChatGPT, Perplexity, Gemini, and Copilot. Answers vary between engines and between runs, so test a set, not a single prompt, and record what each engine names.
• Mark where you are invisible. For each query, note whether the AI names you, cites your page, or names a competitor and skips you entirely. The queries where a competitor wins and you are absent are your loss list.
• Attach a dollar value. Multiply the share of buying-intent queries you lose by the volume of buyers asking and by what one customer is worth. That product is the cost of invisibility for that category, today.

One caution on step two. Because AI answers are non-deterministic, a single check is a vanity metric. The same prompt run twice can return different brands and different sources. A credible cost figure needs a panel of queries across engines and runs, not one screenshot. We go deeper on why a single check misleads in our piece on why your website is invisible to AI.

A worked example, clearly an illustration

The numbers below are made up to show the arithmetic. They are not data, not benchmarks, and not a claim about any real business. Swap in your own figures and the method holds.

Say a regional B2B services firm has 50 core buying queries in its category. It runs them across the four engines and finds the AI names a competitor and skips the firm on 30 of the 50. The firm closes roughly 40 new clients a year from this category, and the lifetime value of a client is about $12,000.

Input (illustrative, not data)	Value
Core buying queries in the category	50
Queries where AI names a competitor and not the firm	30 (60% invisible)
New clients won from this category per year	40
Lifetime value per client	$12,000
Annual revenue exposed to AI buying queries	40 x $12,000 = $480,000
Share of those queries the firm is invisible on	60%
Illustrative cost of invisibility (annual)	$288,000

That $288,000 is not a promise of recovered revenue, and it is deliberately conservative, because it only counts demand the firm already wins and ignores buyers who never reach it. Notice how different the conversation feels, though. “We might be missing some AI traffic” gets nodded at and forgotten. “We can see AI sending an estimated $288,000 of our own buying queries to competitors, and here are the exact 30 queries” gets a follow-up meeting. Same channel, very different reception, because one is a vibe and the other is a ledger.

Where BlueJar fits, and where it deliberately does not

BlueJar runs this method as a structured audit instead of a manual spot check. It tests 400+ prompts across ChatGPT, Perplexity, Gemini, and Copilot, built from a matrix of buying zones, query types, and funnel stages, then classifies every result so you can see which buying queries you lose and to whom.

That classification matters more than a single score, because a blended visibility number hides the lever. There are two very different failure modes, and they need opposite fixes:

• Cited but not mentioned. Your page sits in the AI’s source list, but your brand never appears in the answer the buyer reads. This is a content-structure problem, and the fix is on your own pages.
• Fully invisible. Neither your domain nor your brand appears, and a competitor wins the answer. This is usually an off-site problem. The fix lives in the third-party kingmaker sources the AI keeps citing: the comparison posts, roundups, and review pages where you are absent and your competitor is present.

A blended percentage cannot tell those two apart, so it cannot tell you where to spend. The per-zone, per-intent view can. If you want the longer argument on why rank is not citation, and why the click you optimized for may never happen, see GEO vs SEO in 2026.

The dollar figure comes from BlueJar’s Lost Opportunity Calculator. It auto-selects a model for your business type, whether Local Service, Auto Dealer, E-Commerce, or SaaS, takes the volume and average order value for each offering, and shows recoverable versus still-lost revenue along with the ROI of acting versus doing nothing. It turns the loss list into the kind of ledger a CFO will sign off on.

What BlueJar will not do is pretend to be an attribution tool. It does not report the clicks AI sends, because that data does not exist in most cases. For tracking what the engines actually say about you over a cadence you choose, mentions and citations rather than clicks, see our playbook on tracking brand mentions in AI search. The honesty is not a weakness in the pitch. It is what makes the rest of the pitch believable.

How agencies sell this without overpromising

If you run an agency, this is the cleanest GEO pitch you have, precisely because it does not ask you to promise traffic you cannot guarantee. You are not selling a forecast. You are showing a client a leak in their own demand and the exact queries causing it. In our partner conversations, the dollar figure is what flips a polite “interesting” into a signed scope.

The structure of the pitch writes itself:

• Run the client’s real buying queries across the engines before the call, so you walk in with evidence, not a deck.
• Show the specific queries where a named competitor wins and the client is absent. Specificity is what lands.
• Convert it to a dollar range with conservative assumptions, and say out loud that it is conservative.
• Scope the fix by failure mode, on-page work for cited-not-mentioned, off-site placement for full invisibility, so the budget maps to the actual problem.

This is the same business-case logic we lay out in our guides on how to pitch GEO services to clients and GEO for digital marketing agencies. The one difference from a typical AI-visibility pitch is the refusal to oversell. You measure the downside honestly, scope the fix to the diagnosis, and let the dollar figure do the persuading.

Want to see which buying queries AI answers with a competitor instead of you? Run your first analysis free at bluejar.ai and get your visibility matrix across ChatGPT, Perplexity, Gemini, and Copilot, plus the Lost Opportunity Calculator that turns the gaps into a dollar figure.