Your CRM can’t see where your buyers made up their minds
A prospect opens ChatGPT and asks which tool she should buy. She gets one answer. Three vendors named, a short paragraph each, done.
She clicks nothing.
Two days later she googles one of those brand names, lands on the pricing page and books a demo. Your CRM writes “direct” in the lead source field, which in most companies is CRM for “no idea, sorry”.
The part that actually decided the deal happened somewhere you have no instrument pointed at. Forrester says 89% of B2B buyers now use generative AI somewhere in the buying process. So this is not an edge case that a smart attribution model will smooth out later. It is the front half of the funnel, and it went dark around the middle of 2025.
Every CRM is built on one assumption: that we can see how people found us. That assumption expired. Nobody sent an email about it.
Five numbers. That is the whole thing.
“AI search visibility” sounds like a category invented by a conference committee. It gets much less mystical once you write down what actually gets measured. There are five numbers, they are all boring, and boring is good because boring fits in a database.
- How often you get named when someone asks a question in your category. Market share, basically, in the channel nobody is measuring.
- How often you get cited, meaning your own page was the source the model used. Not the same thing as being named, and the gap is embarrassing. Semrush ran 126 million US prompts in early 2026 and found that on Gemini, the overlap between brands mentioned and domains cited drops as low as 30%.
- Who shows up next to you. Every answer is a shortlist. You are on it with someone.
- How the model talks about you. Not whether you appear, but in what tone. An LLM repeating a bad 2024 review to ten thousand buyers, with no right of reply, is a customer success problem in a marketing disguise.
- Which AI crawlers read your site. GPTBot, ClaudeBot, PerplexityBot. They are already in your server logs. Nobody is looking at the logs.
Five numbers, refreshed weekly, split by country and engine. That is the data.
Why these numbers belong in the CRM
Put those five numbers in a separate dashboard and you have a report nobody opens. The SEO team glances at it on Mondays, sales never sees it, and customer success doesn’t know it exists.
Put them on the account record and they start answering questions people already ask every week.
Attribution finally gets an input. Not a click path, to be clear. Anyone promising you deterministic AI attribution is selling you a story. But when mentions for “best vendor for X” climb in Germany and German direct traffic climbs two weeks later, you can stop shrugging in the board meeting.
Positioning gets audited by an outsider. If the models recommend you for mid-market logistics and never once for enterprise manufacturing, then something other than your positioning deck has decided what you are. That finding does not belong in an SEO tab three menus deep. It belongs next to the ICP definition, where someone who owns targeting will actually see it.
Sales gets a battle card that refreshes itself. For the exact questions a buying committee asks, you see who else is in the answer and what the model read to get there. More useful than the deck a product marketer last touched in March, I would argue.
And customer success gets an early warning system, which I honestly think is the most underrated one on the list. The renewal dates are already in the CRM. The reputation data should be sitting right next to them. If the models start describing your product as “hard to implement” three months before a big renewal, you want to know before the customer’s new CFO asks ChatGPT for alternatives.
It works in the other direction too
This is the part that gets missed. The CRM doesn’t just display visibility data, it makes the data better.
Your closed-lost reasons tell you which competitors to track. Your pipeline by country tells you which markets and languages to monitor. The questions prospects ask in discovery calls are close to the prompts they typed into ChatGPT a week earlier. A prompt set built from real CRM data measures your actual market. A prompt set built from a keyword tool measures somebody’s guess about it.
So the two belong together because each one fixes the other’s blind spot. The CRM knows who bought and why. AI search data knows what they were told before they ever talked to you.
How a “How did you hear about us?” form fixes AI search attribution
Most attribution setups were built for a world where people click a link and land on your site. AI search doesn’t work that way. Someone asks ChatGPT or Claude for the best tools in your category, reads your name in the answer, and then types your URL into the browser days later, or Googles your brand. Your analytics tool records that visit as “direct” or “organic branded,” and the AI engine that actually created the demand gets no credit.
The simplest fix is also the oldest one in marketing: ask people. A single “How did you first hear about us?” field on your signup or demo form captures what analytics can’t see, which is where the buyer’s journey started in their own head.
Why self-reported attribution works for AI search
AI search behaves like a brand channel, not a click channel. It builds awareness and trust inside a conversation, and the conversion happens somewhere else. Last-touch attribution will always undercount it, for the same reason it undercounts podcasts, PR or a conference talk.
At OtterlyAI, the gap is large enough to change strategy. When new signups were asked where they first heard about us, Claude and ChatGPT together influenced 25% of net new MRR in June, more than double Google’s 11%. Claude-sourced signups converted to paid at 9.5%, compared with 5.0% for Google and 4.2% for ChatGPT. None of that would be visible from referral traffic alone.
The form also reveals something traffic data hides completely: which channels bring customers rather than just signups. Google brought in 16% of our signups but only 11% of revenue. Claude brought in 10% of signups and 18% of revenue. Without a self-reported field tied to revenue, you would invest in the wrong channel.
How to set up the form
Keep the question short and put it where almost everyone passes through: your signup form, trial registration or demo request form. A few design choices make the data far more useful.
Use a dropdown with AI engines listed separately. Offer ChatGPT, Claude, Perplexity, Gemini, Google AI Overviews / AI Mode and Microsoft Copilot as individual options next to Google Search, LinkedIn, word of mouth, events, podcasts and so on. A generic “AI tool” option throws away the most valuable part of the data, because each engine sends very different buyers.
Add an “Other” option with a free-text field. People will tell you things you didn’t think to list, such as a specific newsletter, a Reddit thread or a colleague’s recommendation. Review these answers monthly and promote recurring ones into the dropdown.
Make it required, but keep it to one question. One mandatory field adds almost no friction. Five optional ones get skipped.
Randomize or alphabetize the options. This reduces the “click the first answer” effect, which is the most common criticism of self-reported data.
Connecting the answer to your CRM
The form is only half the job. The answer has to travel with the contact through your funnel so you can connect it to revenue, not just signups.
- Store it as a dedicated contact property. In HubSpot or a similar CRM, create a property such as “Self-reported source” and map the form field to it. Set it so the first answer is never overwritten by later form submissions.
- Carry it to the deal or subscription. Use a workflow to copy the value onto the deal, or join it to your billing data (for example, Stripe) through the customer email. This lets you report on MRR, conversion rate and average contract value by self-reported source.
- Keep the analytics source next to it. Most CRMs already record an original traffic source. Keep both fields and report on them side by side. When “direct” in your analytics lines up with “Claude” or “ChatGPT” in the form, you have found your hidden AI search demand.
- Build one monthly report. Show signups, paid conversions, net new MRR and conversion rate for each self-reported source. Track the trend over time rather than overreacting to a single month.
Pairing attribution with visibility data
Self-reported attribution tells you that AI search converts. It doesn’t tell you why. That’s where AI search visibility monitoring comes in. If Claude drives your best customers, you want to know which prompts you appear in on Claude, which competitors appear next to you, and which sources the engine cites when it recommends you. Put the two data sets together and you can see how changes in visibility show up in pipeline and revenue a few weeks later.
Common pitfalls
A few mistakes come up again and again. Grouping every AI engine into one bucket hides the differences that matter most. Measuring signups instead of revenue makes high-volume, low-value channels look better than they are. And treating the form as the only source of truth is risky too, since people misremember. The data is most reliable when you read it alongside your CRM’s traffic source and your visibility metrics, and look for patterns that hold across months.
A single form field won’t give you perfect attribution. But for a channel that barely shows up in analytics, it’s the most direct way to find out whether AI search is actually bringing you customers.
The market has noticed
On 31 October 2025, HubSpot bought XFunnel, a company about ten months old. In the announcement HubSpot mentioned that leads arriving through AI convert roughly three times better than leads from traditional search. Wonderful if you sell lead scoring. Slightly terrifying if you think about it for more than a minute, because your best cohort has your worst data attached to it.
On 28 April 2026, Adobe closed its acquisition of Semrush, announced at about 1.9 billion dollars. Adobe’s own client data showed AI-referred traffic to US retail sites up 269% year over year as of March 2026.
(Vendor numbers from vendor networks, so squint at the decimals. The direction is not in dispute.)
Two very different companies reached the same conclusion: this data belongs next to customer data, not in a silo.
And then there is Europe 🇪🇺
Once AI visibility data lives inside the CRM, the vendor behind it is no longer a marketing tool on the side. It sits in the same system as your customers’ contact records, deal notes and renewal data. That changes who you should be comfortable working with.
Frankfurt is a city, not a jurisdiction. Keeping data in a German data centre does not put it out of reach of the US CLOUD Act if the provider is a US company, and GDPR Article 48 gives an EU controller no clean way to comply with such an order. That used to be a conference panel argument. It is now line 14 of the procurement form.
It got sharper fast. The EU Data Act became fully applicable in September 2025. The Cloud Sovereignty Framework arrived in October 2025 with actual scoring. The Commission’s June 2026 sovereignty package pushed the question down the supply chain, so a company running on a sovereign cloud still has to account for the US SaaS its sales team opens every morning. CRM and marketing automation are named explicitly.
For a European company, the practical answer is simple: when a tool connects to your CRM, pick a vendor that is incorporated in the EU, operates under EU law and has EU sub-processor terms. That is how you keep your customers’ data protected under GDPR rather than just stored in the EU. Ask where the vendor is incorporated, who its sub-processors are, and whether any non-EU authority can compel access. If the answers are vague, that is your answer.
The second European difference has nothing to do with law. AI answers are local. A German question returns German sources and German competitors. Engine availability still varies across the bloc. A prompt set built on US English tells a French insurer approximately nothing about its own market. Monitoring has to run in the languages and countries where your customers actually buy.
(Disclaimer: I co-founded OtterlyAI, which is built and run in Vienna under EU law, so read this section with the appropriate squint 🦦)
The question for this year
The first half of your funnel moved into AI answers. Your CRM still only sees the second half. The fix is not a smarter attribution model. It is putting the five numbers where decisions about accounts, positioning and renewals are already made.
So, honest question for anyone running a CRM: is AI search data anywhere near your account records yet, or is it still filed under “SEO, someone else’s problem”? 👇
