- Services
-
INSIGHTS, PLAYBOOKS & GROWTH STORIES
Subscribe to our newsletter
- Learn
How do you measure the real impact of a channel that produces almost no clicks? And how do you defend organic growth budget to a board that only sees traffic falling? In this 26 August webinar, Jake Stainer, explains why this linear, click-based thinking is no longer valid in AI-first search, and how to attribute the actual influence this channel has on your revenue targets.
Direct isn’t a channel, and you shouldn’t be measuring it. It’s simply the bucket an analytics tool uses when it cannot identify a source. Reporting on it as a channel tells you nothing
The buyer journey has changed and is no longer linear and click-based.
That’s why direct has been climbing for years.
GA4’s AI Assistant isn’t the answer to attribution in the AI-search era, because it only records people who clicked through immediately from an AI search result.
But AI chats rarely produce a click. Most buyers read the answer, remember the brand, and return later. AI Assistant records none of this.
Citations are valuable, but they’re not the end goal. An AI search engine can cite your content as a source while excluding you from the answer. It may even recommend your competitor.
Since so much of the buyer journey is no longer measurable, you need a hybrid attribution model that combines software attribution (what is measurable) with self-reported attribution (asking buyers directly).
Begin by tracking this leading indicator: brand coverage. Once your AI presence is established, start analyzing share of voice.
Brand coverage is foundational: Do AI search engines mention you and describe you correctly?
Share of voice guages how often you appear when compared to chosen competitors.
This is where the hybrid attribution model becomes particularly important. Branded search and direct traffic are essentially harvesting the combined impact of multiple marketing activities. GEO, PR, word of mouth, social, paid campaigns and other activities can all increase those numbers.
From traffic data alone, there isn’t an easy way to cleanly separate them. You could run isolated geographic experiments. For example, invest heavily in paid media activity within a specific city or region and compare the subsequent movement in branded search and direct traffic against a control region. That can give you a stronger causal signal, but it can also be expensive and impractical.
For most companies, the more realistic solution is combining software attribution with self-reported attribution.
Software attribution captures what is observable. SRA helps explain the parts of the customer journey that software can’t see. There is always criticism that SRA isn’t perfectly accurate. People forget things, there may have been multiple touchpoints, and the answer they give might not represent every interaction they had with your brand. But I don’t think that’s a reason to dismiss it.
If somebody tells you ChatGPT, a podcast, a colleague or a particular campaign was where they discovered you, or was the most influential source in their journey, that’s extremely useful information.
More importantly, attribution shouldn’t be treated as an exact science. The objective isn’t to take your attribution report, calculate that 17.4% of conversions came from one channel, and therefore allocate exactly 17.4% of your budget there. Attribution is one source of evidence for making better bets.
You combine attribution data with customer insight, market understanding, creative thinking and experimentation to decide where you have the highest confidence in investing next.
AI search visibility itself can be measured almost in real time. If your AI visibility platform is tracking prompts every day, you can look at a seven-day window and see whether your visibility, citations or share of voice are increasing.
The harder question is: How long should it take before increased visibility translates into demos, pipeline and revenue? There isn’t one universal lag because it depends heavily on what you’re measuring.
The first thing I would look at is the prompts themselves. Are you gaining visibility for commercially relevant prompts used by your actual ICP, or are you simply becoming more visible overall?
If visibility is increasing but demos and leads aren’t moving, I would diagnose the chain between those two metrics. I’d look at:
Once those foundations are aligned, you can start looking for the correlation between increased visibility and downstream commercial outcomes.
So if your CMO wants revenue this quarter, I wouldn’t simply promise that a specific increase in visibility will produce a specific amount of revenue within a fixed number of days.
I would make sure you’re increasing visibility for the right prompts, in front of the right ICP, with the right positioning, and then measure how that change flows through to demos, pipeline and eventually revenue.
I think you can build a forecasting model, but initially I would treat it as a rough hypothesis rather than a precise prediction.
Before forecasting anything, you need to make sure the fundamentals from the previous question are aligned: ICP, prompt selection, funnel stage, positioning and the wider customer journey. Once those are in place, you can create a simple initial hypothesis.
For example: If we double commercially relevant AI search visibility, what happens to demos?
You can forecast an expected relationship and then measure what actually happens. Perhaps you initially assume that doubling visibility should double demos. Over time, you’ll see whether those metrics increase in step. Maybe demos only increase by 40%, or perhaps certain types of prompts produce significantly more pipeline than others.
The difference between your forecast and the actual result gives you a margin of error. As you collect more data, you can continuously improve your model. Eventually, you can start connecting the entire chain:
That allows you to move from simply reporting visibility metrics to estimating the commercial return of additional investment in AI search.
I wouldn’t expect that model to be highly accurate on day one. The value comes from building the baseline, measuring what actually happens, and improving the forecast as your own historical data grows.
The way I would answer this is through incrementality testing. The fundamental question is simple: If this advertising didn’t exist, how many of these conversions would we have received anyway?
You can test this in different ways. You might compare two groups, run a controlled experiment, or even compare periods where a particular campaign is switched on versus switched off.
Branded paid search is a good example. You could switch off branded paid search for a period and measure what happens to total conversions. Do those conversions disappear, or do people simply click the organic result instead?
The difference between what happens with and without the advertising is the incremental value you’re actually buying. This is important, because platform-reported cost per acquisition (CPA) can be extremely misleading.
Imagine branded paid search reports a CPA of $100, but your incrementality study finds that only 20% of those conversions were actually incremental. In reality, you’re effectively paying around $500 for each incremental conversion. The other 80% may have happened regardless.
So rather than asking, “What CPA is Google Ads reporting?”, I would ask: What is our cost per incremental conversion? That’s a much more useful number for deciding whether your paid budget is genuinely creating additional growth.
Yes, SRA is the answer, and this is exactly why I recommend moving away from last-click attribution and looking at first-click attribution combined with self-reported attribution (SRA).
The fundamental problem with last click is that it over-indexes on channels like branded paid search and branded organic search. These channels are often capturing demand that already exists rather than creating new demand. That can seriously distort how you allocate budget. You end up investing more money in defending existing branded real estate rather than reaching new audiences and creating future demand.
I would start with first-click attribution for software attribution. This gives you a much better indication of which channels are actually introducing new people to your brand. Then combine that with SRA.
For example, paid search is relatively easy to identify because it’s click-based and carries UTMs. I would bucket those leads first, while excluding branded keywords wherever possible. Again, the goal is to avoid giving excessive credit to someone searching specifically for your brand.
You can then use SRA to distinguish between channels such as traditional organic search, AI search and other sources of discovery.
The result is a hybrid attribution model. Software attribution tells you what you can reliably observe through digital click-based behavior, while SRA helps fill in the demand-generation activity that traditional attribution struggles to see.
That gives you a much better basis for investing in channels that create demand and reach new audiences, rather than simply rewarding the channels that happen to capture it at the end.
