Most marketing dashboards now measure the last few steps of a decision that was made somewhere else. A buyer asks ChatGPT for a shortlist, a colleague drops a name in a group chat, and by the time anyone fills in your contact form, the choice is half made. You can’t track those rooms. You can measure what comes out of them, and that takes three changes: ask buyers where they heard of you, count the people in a buying group you reached before a deal existed, and judge channels as a system over months instead of one at a time.
Your dashboard sees the end of the decision
The private part of buying has grown faster than the tools built to watch it. In LinkedIn research presented at Advertising Week Europe and [reported by Marketing Week]( https://www.marketingweek.com/b2b-marketing-metrics-linkedin/), 94% of B2B buying groups use large language models such as ChatGPT or Gemini before they contact sales. A separate analysis of more than 50,000 closed deals at 850 B2B companies, run by Factors.ai and shared by LinkedIn, found that buying groups explore solutions for about 124 days before they speak to a seller, [according to Social Media Today]( https://www.socialmediatoday.com/news/linkedin-shares-insights-into-key-b2b-buying-behaviors/831324/)&source=gmail&ust=1793349936483000&sa=E.
Four months is a long time to be invisible to your analytics.
Consumers are moving the same way. Kantar’s [Media Reactions 2026]( https://www.kantar.com/inspiration/advertising-media/media-reactions-2026) study of about 23,000 consumers in 33 markets found that 73% use AI assistants and a third already research brands and products through them. Only 24% trust the brand recommendations those assistants give. That gap matters for measurement. People ask the machine, then check with someone they trust before they act, and both steps happen off your dashboard.
Marketers feel it. In the same study, [Kantar reported]( https://www.kantar.com/press-center/net-75-percent-of-marketers-plan-to-increase-investment-in-ai-visibility) that only 57% of advertisers are confident they have the right media mix, down from 64% a year earlier. Half are confident in their balance between brand building and performance marketing, down from 60% in 2025. Just 51% say they are getting useful intelligence out of their data.
Nearly half of what works is channels working together
Here is the number most measurement plans ignore. Kantar’s LIFT+ database shows that between 2020 and 2025, 43% of campaign impact came from channels working together, up from 18% before 2014. A dashboard that credits each channel on its own misses almost half the effect.
Jeff Greenspoon, Kantar’s CEO for the Americas, told [Marketing Dive]( https://www.marketingdive.com/news/how-marketers-can-optimize-their-media-mix-as-confidence-takes-a-turn/833197/) that teams with separate dashboards and partners end up “measuring the instruments and not the orchestra.” Each instrument reports a tidy number. None of them describes how a buyer came to trust you: a film they watched on YouTube, a founder’s post a colleague forwarded, an AI answer that named you, a website that confirmed what they had heard.
Kantar’s Gonca Bubani argued in the press release that much of the falling confidence is unjustified for this reason. The media is working. The way it gets measured has not caught up.
The AI leads already in your CRM, filed under “direct”
If you believe AI assistants are not sending you business yet, ask how you would know. David Khim, writing in [Search Engine Journal]( https://www.searchenginejournal.com/ai-visibility-measurement-what-to-track-what-to-ignore/582009/), reported that in his firm’s HubSpot data, 80% to 90% of leads from AI platforms were tagged as “organic” or “direct.” The cause is mundane. People who get an answer from ChatGPT rarely click the link; they search your name or type your address, and analytics files the visit as if it came from nowhere.
Khim also describes a client that found about 5% of its registrations came from ChatGPT before it had done any AI visibility work at all. That share was invisible until someone asked.
This is the cheapest insight in measurement right now. One field on your form will tell you more about AI’s role in your pipeline than any referral report.
Measure the buying group instead of the click
The LinkedIn and Factors.ai analysis points to a better unit of measurement. Reaching six or more contacts at an account before a deal is ever created in the CRM lifted win rates by 17 percentage points. Engaging three or more additional contacts during the active cycle lifted them by 16 points. Companies that paired LinkedIn ads with organic LinkedIn activity saw win rates 2.4 times higher than those running paid campaigns alone.
The same analysis found that 76% of LinkedIn ad spend targets director-level roles. That is a bet on one person in a decision that belongs to a group. LinkedIn’s earlier research, in the Marketing Week report, found brands are more than 20 times more likely to be bought when everyone in the buying group already knows them, and that buyers favour the choice they could defend if something went wrong.
Put those findings side by side and the metric changes. The useful question is how many of the people who will sign off on this decision had already heard of you, and from whom.
This is where story and craft show up in the numbers. A buying group of six people will not all see your ad. They will see what one of them sends to the others: a two-minute film of your team at work, a case page with real detail, a founder’s post that explains a hard decision plainly. Material that gets forwarded reaches the people your targeting missed, and most companies keep the work that would travel this way hidden.
A measurement system a growing company can run this quarter
You do not need a media mix model to start, and most growing companies lack the years of spend and sales data that would make one reliable. Here is what we would put in place first.
- Add one question to every form and first call. Ask “How did you hear about us?” as an open field, with an option for AI assistants such as ChatGPT or Gemini. Khim suggests testing the field’s effect on conversion, and having sales ask on discovery calls for deals that skip the form. Log every answer in the CRM next to the deal value.
- Count named people per account before the deal. For your ten most important target accounts, record how many people in the likely buying group engaged with you in the past 90 days: watched a film, opened an email, followed the founder, came to an event. Then watch whether wider exposure means faster closes.
- Keep a list of the prompts your buyers actually type. Pull the wording from sales call recordings and onboarding conversations, as Khim recommends, and check monthly whether AI assistants name you. Treat it as a monitor.
- Review channels as a portfolio every quarter. When a brand film went live, did branded search and self-reported mentions rise over the following weeks? That is how a small team hears the orchestra.
- Report pipeline and revenue first. Leadership should see deals, then the signals behind them.
Give it a quarter, then read the answers together. When several new leads write some version of “a colleague sent me your video” or “ChatGPT suggested you,” you have learned something no referral report would show, and you know which assets to make more of.
What to watch, and what not to turn into a target
Some numbers deserve attention without deserving a target. Khim cites research showing that even for branded queries, 48% of the sources AI tools cite are earned media, 30% are other companies’ commercial content, and only 22% are the brand’s own site. That tells you where reputation is built. It does not make citation counts a KPI.
AI referral traffic is even less stable. Khim notes that OpenAI cut the number of sources ChatGPT displayed, referral traffic fell for many sites, and about a month later the change was reversed. A KPI that moves because another company changed its product is a weather report.
There is an honest limitation to everything above. Self-reported attribution is messy, because people forget or simply name the last thing they remember. The LinkedIn figures come from a company that sells LinkedIn ads, so read them as directional. None of these methods produces the clean, single number a finance director wants. They produce a truer picture, and a rough true picture beats a precise wrong one.
What this means for 2027 and 2028
This part is prediction. Kantar found a net 75% of marketers plan to increase investment in AI assistant visibility in 2027, including paid ads. As money moves there, the pressure to prove a return will grow, and the companies that can answer “where did this deal come from” in a buyer’s own words will make better calls than the ones reading referral reports.
We expect the private share of buying to keep growing. That moves the weight of marketing toward what buyers repeat to each other and to machines: a clear story, documented proof of the work, a founder whose thinking gets forwarded, a website that confirms what people heard. Those are the assets that show up in the “how did you hear about us” field, and they are the hardest for a competitor to copy with the same AI tools. At Revolt, that is why we build the film and the website around one story and ask clients to measure them as one.
If your marketing gives people nothing worth repeating, no dashboard will find it. The next step is a single form field and a quarterly review, and if you want help mapping where your buyers actually hear about you, start a consultation with us.
