How to Measure ROI From Zero-Click Searches

The click alone can't do this job anymore. About 68% of US Google searches ended without a click in early 2026, by SparkToro's count. If your ROI model only credits visits, it ignores most of the moments when a buyer met your brand. Our guide to the headless brand explains why so much marketing now happens off your site; this post is the measurement half.

You'll get five signals, a simple model we call the Zero-Click ROI Ledger, and a worked example with real arithmetic. For the studies behind the 68%, see what the data shows on zero-click searches.

Why Click-Based ROI Misses Zero-Click Searches

A click report measures one path: search, click, convert. Zero-click searches break the first link. The buyer reads your name in an AI Overview, a ChatGPT answer or a featured snippet, closes the tab, and comes back days later by typing your URL or searching your brand. Your analytics tool records a direct or branded visit. It never learns what started it.

Seer Interactive caught this in its own data. In October 2025, on informational queries where a tracked brand was cited in the AI Overview, impressions more than doubled, from 15.8 million to 33.1 million, while clicks held flat at about 400,000. Click-through rate fell by half. Read alone, that looks like failure. Read with impressions, it may mean the brands were cited on more queries, though Seer says it can't confirm the cause without account-level data. Its advice: "Always look at clicks and impressions separately."

So the fix isn't a better attribution tool. It's a wider set of signals and an honest model. Bain's zero-click report makes the same call, telling marketers to shift from click-focused metrics toward search impressions and AI reach.

Five Signals That Show the Value of Zero-Click Searches

Years before AI Overviews, SparkToro's Rand Fishkin argued that most hard-to-measure marketing shows up through three doors: direct visits, branded search and higher conversion rates. Zero-click searches fit that pattern well. Add visibility at the top and AI referrals at the bottom, and you have five signals.

1. Impressions and share of voice

This is the exposure that zero-click searches create. In Search Console, the generative AI performance report counts how often your pages appeared in AI Overviews and AI Mode, by page, country and device. Google rolled it out to every site by 31 August 2026. It shows impressions only, not clicks.

Bing Webmaster Tools goes further for Copilot and Bing. Its AI Performance report counts citations and cited pages, and since June 2026 it adds Citation Share for each grounding query. For ChatGPT, Perplexity and Claude, you need a prompt panel: a fixed list of buyer questions you run on a schedule. Your AI share of voice is your share of all brand mentions across those answers. Our guide to measuring GEO covers panel size and run counts.

Some teams put a dollar figure on impressions by pricing them like ad reach. That's a media-value estimate, not a return. Keep it out of the ROI line, or label it clearly.

2. Branded search lift

When answers name you, some readers search for you later. Search Console's branded queries filter, available to all eligible sites since 11 March 2026, splits your clicks and impressions into branded and non-branded. Google classifies them with an AI-assisted system that catches misspellings and product names, and it warns that some queries may be misidentified. It works only on top-level properties with enough search volume.

Compare branded clicks with their own trend, not just last quarter. If branded search grew 4% a quarter before your program, only the growth above that line is lift. Check for a launch, a price change or an ad campaign in the same weeks, since each can move branded search on its own.

3. Direct traffic

GA4 defines Direct as visits from a saved link or a typed URL. In practice it also collects visits whose referrer got lost along the way, such as links opened from some apps or pasted into a new tab. That makes Direct a useful supporting signal and a poor primary one. Watch it for a rise that lines up with your visibility gains, but don't give it its own line in the ROI model, because the same visitors often appear in branded search or your form data.

4. Self-reported attribution

Ask people. Add a required "How did you hear about us?" field to your signup or demo form, with an open text box or options that include "ChatGPT or another AI assistant" and "Google's AI answer." It's the only signal that lets a buyer who never clicked tell you so directly.

Ahrefs does this at scale. It tracks signups that credit AI at onboarding, and in March 2026 ChatGPT was named by 1,978 new users, Claude by 2,836 and Gemini by 619. Self-reports are imperfect: people forget, and the answer options you offer can steer them. They're still the closest thing to a direct measure of zero-click influence.

5. AI referral traffic

These are the clicks that did happen. GA4's AI Assistant channel covers sources like ChatGPT, Gemini, DeepSeek, Copilot and Grok, but not Google's AI Overviews and AI Mode, which stay in Organic Search. A custom channel group catches the assistants Google leaves out. Our GA4 setup for AI referral traffic walks through it.

Expect many of these visits to land on your homepage. SE Ranking found 60% of AI-referred visits land on homepages, against 17% from organic search. Judge the channel as a whole, not by which article it reaches.

The Zero-Click ROI Ledger

The ledger is a one-page model that turns the five signals into a return. It has four rules.

1. Count incremental outcomes only. Use the change from a baseline period, adjusted for any trend that was already under way.
2. Label each line. Measured means you counted it (signups from the AI channel). Estimated means you derived it from a measured signal (branded lift times a conversion rate). Assumed means you chose it (customer value).
3. Never add overlapping signals. Self-reported AI attribution and branded search lift often describe the same buyers. Put the smaller one in the low case and the larger one in the high case.
4. Report a range and the break-even point. A range admits the uncertainty. Break-even tells leaders how much has to be true for the program to pay.

The formula is simple:

ROI = (new customers × value per customer − program cost) ÷ program cost

New customers come from three lines: extra clicks, extra AI referral signups and no-click influence. The first two are measured. The third is the one zero-click searches add, and it's where the range comes from.

Worked Example: Tallyfold's First Quarter

Tallyfold is a fictional invoicing and payments app for agencies. All numbers are illustrative. Last quarter it spent $12,000 on 30 new answer-first pages and refreshes, aimed at question and comparison queries, the kind most exposed to zero-click searches.

The signals, quarter over quarter

Impressions rose far faster than clicks, the same pattern Seer saw. A 12-point share of voice gain clears the roughly 9-point bar our GEO measurement guide gives for a rate measured on 180 answers, so it's likely real.

The ledger

Tallyfold converts 30% of signups to paying customers, which gives 14.4 to 23.1 new customers. It assumes each customer pays $52 a month on average (the $39 base plan plus extra users) and stays 18 months, a value of $936.

• Low case: 14.4 × $936 = $13,478. ROI = ($13,478 − $12,000) ÷ $12,000 = 12%.
• High case: 23.1 × $936 = $21,622. ROI = ($21,622 − $12,000) ÷ $12,000 = 80%.
• Break-even: $12,000 ÷ $936 = 12.8 customers, or about 43 signups at a 30% conversion rate.

The report to leadership reads: the program returned between 12% and 80% on 18-month customer value, and it would have broken even at 43 signups. The low case, 48 signups, only just clears that bar, which is exactly what leadership should know.

What changes the answer

Change the value assumption to first-year revenue ($52 × 12 = $624) and the same quarter returns −25% in the low case and +20% in the high case. Nothing about the marketing changed; only the horizon did. Agree on the horizon before you report, and say which one you used.

Notice what the ledger leaves out. The 210,000 extra impressions and the 1,600 extra direct sessions aren't in the dollar line. They support the story that visibility grew, but valuing them too would count the same buyers twice.

How to Show the Program Caused the Lift

A ledger shows that value arrived. It doesn't prove your pages caused it. Three methods help, from simplest to most demanding.

• Changed vs. untouched topics. Split your target questions into ones you worked on and similar ones you left alone. If visibility and branded interest rise only for the first group, your pages are the likely cause. The control-prompt test in our GEO guide shows the math.
• A correlation dashboard. Chart impressions, share of voice, branded search, direct visits and signups by month on one page, and note every launch, campaign and outage. Fishkin's advice is to ask leaders for at least three months, and ideally six to twelve, before judging.
• Geo experiments and media mix models. Larger teams can test by region or model all channels together. Google's open-source Meridian supports media mix modeling and geo experiments, and it can take in search query volume data to control for organic demand.

Whichever you use, keep your ROI claim for zero-click searches modest. "Between 12% and 80%, with self-reported data as the floor" is more believable than a single precise number.

Where Rankbox Fits in Zero-Click Measurement

Rankbox is on the cost side of this ledger, not the measurement side. It researches the questions your buyers ask and writes source-backed, answer-first articles with its Citation-Ready Writer, which is the kind of work that earns impressions and mentions in AI answers. Rankbox doesn't track AI citations, branded search or referral traffic, so the signals above come from Search Console, Bing Webmaster Tools, GA4 and your own form data. For the prompt panel, our free AI Visibility Prompt Kit writes 30 buyer prompts with a scorecard. See pricing if you want Rankbox on the cost line.