How to Instrument Your CRM to Report AI-Sourced Pipeline

Most AEO measurement advice ends at a GA4 segment for AI referrer domains and a "How did you hear about us?" field. Then the quarter ends, your VP of Sales asks how much pipeline ChatGPT or Claude actually produced, and you discover the reporting was never the hard part. The CRM was.

The GA4 layer is solved and well documented. If you have not built it yet, start with tracking AI referral traffic in GA4 and come back. This page is about what happens after the session lands: where the signal goes, which object holds it in HubSpot and in Salesforce, what breaks it, and how to report it without overclaiming. The HubSpot and Salesforce steps reflect each vendor's documentation as of October 2026.

How do you attribute a demo that started with a Claude recommendation?

You attribute it in two layers: a source value for the buyers who clicked through from Claude, and a deal-level influence record for the much larger group who read the answer, closed the tab and came back later. Most teams build only the first layer and then wonder why the AI line in their pipeline report stays near zero.

The click path is the easy one. Claude cites your page, the buyer clicks the citation, lands on your site and books a demo in the same visit or a later one. HubSpot records that session with AI Referrals as the traffic source and the AI platform domain, such as claude.ai, in Drill-Down 1. Salesforce records nothing on its own, so you capture the referrer and any UTM values in hidden form fields and write them to the lead.

The no-click path is the common one. The buyer asks Claude for a shortlist, sees your name, and a few days later types your brand into Google or straight into the address bar. Every system you own logs that visit as branded organic or direct, and only the buyer can tell you otherwise. That is why the self-reported field and the sales call carry so much weight below, and why our pipeline and ROI model for AI discovery treats influence, not origin, as the unit.

The further down the table, the more common the path, and the less your CRM knows about it.

Why does my CRM show AI-sourced demos as direct or organic?

Your CRM shows AI-sourced demos as direct or organic because it can only record the last trackable touch, and in AI discovery the deciding touch happens inside a chat window you cannot see. A buyer asks Claude for tools in your category. Claude names you. The buyer does not click. A few days later they search your brand name, land on your homepage, and book a demo.

Every system in your stack records that correctly and still gets the answer wrong, because the recorded touch is the last step of a decision already made inside an AI answer.

The share of buyers on that path is already large. In G2's 2026 Buyer Behavior Report, published in July 2026, 82% of buyers said they had sourced software recommendations from an AI chatbot in the last 24 months. Most of your market may have built its shortlist somewhere your attribution model cannot see.

So origin attribution is the wrong goal. You won't prove a deal originated in Claude, because the buyer's path routes around your instrumentation. What you can prove is that AI influenced a specific set of deals, at a specific rate, with a documented margin of error. That is a weaker claim than most AEO vendors make and a much stronger one than your CFO will get anywhere else. The related problem of proving pipeline when the buyer leaves no referrer is the same problem viewed from the traffic side.

What does HubSpot record on its own?

HubSpot records AI Referrals as a native value on its Original and Latest Traffic Source properties, with the AI platform domain in Drill-Down 1 and the utm_campaign value in Drill-Down 2. If you are on HubSpot, you are already capturing the click path and probably not reporting on it.

The drill-down rules that wreck data

Two drill-down rules in HubSpot can wipe out a quarter of data before anyone notices.

First, according to HubSpot's documentation on source properties, the drill-down properties are set automatically and cannot be updated manually, so no ops team can backfill them by hand.

Worse, if you change a record's traffic source, the drill-down properties are cleared. A well-meaning rep who "corrects" a contact from Direct to AI Referrals destroys the underlying detail permanently. Lock the property or train against it before you roll anything out.

Getting AI Referrals into attribution reports

HubSpot's attribution reports group credit by interaction source, which is the traffic source of the session where each interaction happened. So a contact who arrived from claude.ai and later booked a meeting carries an AI Referrals interaction into the report without any custom work. A custom checkbox you invent does not become an interaction source, so keep your custom fields for the parts the native source cannot express.

Check your tier before you promise anyone a dashboard. Contact create attribution reports need Marketing Hub Professional or Enterprise, while deal create and revenue attribution reports need Marketing Hub Enterprise. On a lower tier, you report AI-sourced pipeline from a deal list filtered on your own AI Influenced property instead.

How do you set this up in Salesforce?

In Salesforce you build the AI source yourself: a Lead Source value, hidden form fields that capture the referrer, and a campaign that Customizable Campaign Influence can credit. It is more work than HubSpot and gives you more control.

Capture the click path at the form

Salesforce does not read the browser's referrer, so your demo form has to. Add hidden fields for the referring domain and the UTM parameters, fill them with a small script on page load, and store them on the lead. When the referring domain is claude.ai, chatgpt.com or another assistant, a Flow sets Lead Source to a value such as AI Assistant and fills an AI Referral Source picklist with the engine.

Make campaign influence count AI deals

Create one campaign for AI answer engines and add a lead or contact as a member whenever the AI Referral Source or the self-reported field names an assistant. Campaign membership is what turns a field value into pipeline credit.

The detail most teams miss is the contact role. Salesforce's Customizable Campaign Influence scans active campaigns for members who are also assigned a contact role on an open opportunity. If the person who read the Claude answer is not a contact role on the opportunity, your AI campaign gets no credit, however well you tagged the lead. Make adding the demo booker as a contact role part of opportunity creation.

Then choose the model on purpose. The default Primary Campaign Source model assigns 100% influence to the single campaign in the opportunity's Primary Campaign Source field, which usually means your paid or event campaign wins and the AI campaign shows nothing. Add a first-touch or even-distribution model alongside it so AI influence shows up while the default model keeps deciding everyone else's credit.

HubSpot vs Salesforce: which setup fits your team

Our stance: build where finance reads the number. If board pipeline comes from HubSpot deals on Marketing Hub Enterprise, use the native source and spend your effort on the deal-level fields; on Professional, report deal pipeline from a filtered deal list. If pipeline lives in Salesforce opportunities, build the capture there even when marketing runs HubSpot, and map the HubSpot AI fields across so there is one version of the truth.

Which CRM fields do you need to track AI chatbot referrals?

You need four fields: an AI Referral Source picklist and a Self-Reported Source field on the contact or lead, plus an AI Influenced checkbox and an AI Influence Evidence note on the deal or opportunity. They work the same way in either CRM. Resist the urge to build a dozen more properties.

That last field is the one teams cut, and cutting it is why AEO reporting dies a few months in. We would not ship the other three without it. When your CFO asks how a deal got flagged AI Influenced and the answer is a paragraph quoting the buyer on a discovery call, the number survives scrutiny. When the answer is "the workflow set it," it does not.

Set AI Influenced as a one-way latch. It flips false to true and never back. Deals pick up AI evidence at unpredictable moments, often on a call weeks into the cycle, and a workflow that recalculates on every property change will erase what you learned.

How reliable is the self-reported field?

Self-reported attribution is noisy enough that you should report it as a range with the method attached, never as a precise share. Every AEO article tells you to add "How did you hear about us?" Almost none tell you how wrong the answers are.

What the testing found

In a 2022 test, Ruler Analytics checked its own demo form against its attribution data across more than 350 leads. Of Ruler's inbound demos in that 2022 study, 72% gave vague detail or none at all. The same 2022 study looked at one Ruler client, where 47% of leads who picked the first option in the list were attributed inaccurately, because people choose the first plausible answer and move on.

Recast came at it from the other end. In Recast's 2023 analysis of the "How did you hear about us?" field, answers lean toward the most recent touch, and nearly 30% of people skip the question when it is optional.

Four design decisions that fix most of it

Randomize picklist order per session. If first-position picks are this unreliable, a fixed list means whatever sits at the top absorbs error from everything below it. Put Claude at the top of a static list and you will manufacture a fictional AEO win.

Make it required. Losing that many responses to an optional field costs you more than the friction does.

Ask at conversion, not during onboarding. The same question asked at different moments produces different answers, and the demo form is when recall is freshest.

Pair every picklist choice with one free-text line. The free text is where "I asked Claude for alternatives and you came up twice" lives. The picklist gives you counts. The free text is your evidence for the deal field. Our guide to building a self-reported attribution survey covers the form itself in more depth.

Report a range

Your raw AI count carries a known error term, so report a low and a high estimate of AI-influenced pipeline with the method underneath. A CFO can act on a range. A single precise share invites a question about its error margin that you cannot answer.

Ruler's 2022 test also found that the form caught 1.4% of leads its attribution software would have missed. Combined, the two methods reached 81% accuracy on lead tracking in that 2022 test. That overlap is the argument for instrumenting both layers.

A worked example: one demo, start to finish

A demo that started in Claude reaches your CRM by one of three paths: the buyer clicks the citation, the buyer searches your brand later, or a colleague books the demo, and a referrer-based dashboard sees only the first. A demand gen director asks Claude which tools help a PLG company report pipeline by channel. Claude's answer names your product and cites your comparison page.

Say she clicks. HubSpot sets the contact's Original Traffic Source to AI Referrals with claude.ai in Drill-Down 1. Your hidden fields capture the referrer, the Flow sets Lead Source to AI Assistant, and the contact joins the AI campaign. When the opportunity opens with her as a contact role, campaign influence credits the AI campaign.

Now say she does not click. She searches your brand later and books a demo. Source says Organic Search. On the form she picks Claude from the randomized list and writes "it came up when I asked for reporting tools." A workflow copies that into AI Influence Evidence, flips AI Influenced to true, and adds her to the AI campaign.

In the third version, her VP books the demo and nothing on the form mentions AI. On the discovery call the director says the shortlist came from Claude. The transcript search catches it, the rep pastes the quote into AI Influence Evidence, and the latch flips. In Salesforce, the rep also adds the director as a contact role on the opportunity and a member of the AI campaign, because campaign influence only credits members who hold a contact role. Skip that and the evidence field says AI while the influence report says nothing.

The same deal reaches your report in all three versions.

Mine sales calls, because that is where AI influence is spoken out loud

Your call recordings are the highest-fidelity AI attribution source you own, and nearly nobody queries them. Buyers say it unprompted: "we shortlisted you off ChatGPT," "Claude kept bringing up your comparison page."

Build a saved search across Gong, Chorus, or Fireflies for engine names, plus phrases like "asked the AI," "the chatbot said," and "AI recommended." Run it weekly. Every hit becomes a deal-level evidence entry, with the quote pasted in verbatim.

This is where the one-way latch pays off: call evidence surfaces well after the contact was stamped, and for large deals it is often the only proof.

What is the real conversion rate of ChatGPT and Claude traffic?

There is no reliable benchmark yet: the most quoted figures, 15.9% for ChatGPT and 5% for Claude, come from a single Seer Interactive client's GA4 data. We would not put a borrowed rate like that in a board deck.

Seer Interactive's 2025 analysis of how AI traffic converts also put Perplexity at 10.5% and Gemini at 3%, against 1.76% for Google organic. Read the sample before you borrow any of it. It is one client's GA4 data from October 1, 2024 to April 30, 2025, roughly 11,000 AI sessions against 14 million Google organic sessions, and AI traffic was 0.07% of organic traffic.

So the honest read is narrow: in one documented case, a very small volume of AI traffic converted at a high rate. That is encouraging and it is not a benchmark. Quote it as an expectation and you have set a target you did not measure and cannot hit on purpose.

Your own instrumented CRM, with its ranges and evidence fields, gives you a smaller number you can defend. Use it, and set it beside the other numbers in the AEO metrics a SaaS board wants to see.

A phased rollout that survives contact with the org

Run the phases in order, because each feeds the next. Most of the effort sits in the build; after that it's a light weekly habit.

Phase one: audit. Owner: marketing ops, plus your Salesforce admin. Effort: an afternoon of admin work on HubSpot, closer to a day on Salesforce. HubSpot users, check whether AI Referrals is populating and whether your tier includes the attribution reports you plan to use. Salesforce users, build the Lead Source value, the AI Referral Source picklist and the AI campaign. Lock the source property against manual edits. Done when nobody can hand-edit the source.

Phase two: ship the fields. Owner: marketing ops with whoever owns the website forms. Effort: the biggest build in the rollout, mostly form script and Flow work on Salesforce. Add the four fields. Randomize the self-reported list, make it required, add the free-text line, and add the hidden referrer fields to every demo and trial form. Do not backfill history; you will be inventing data. Done when a test submission from a Claude link fills every field it should.

Phase three: evidence. Owner: sales ops, with reps doing the daily part. Effort: small to set up, then a weekly review of the call search. Turn on the transcript search, start writing evidence entries, and make the demo booker a contact role on every new opportunity. Expect the first month to look thin; thin is accurate. Done when every AI Influenced deal has a note a stranger could audit.

Phase four: report and reconcile. Owner: the marketing lead, reviewed with finance. Effort: a few hours each reporting cycle. Publish the first report as a range with the method attached. Then pull the AI Influenced deals and check them against the citations you are actually earning. If deals are flagged for engines where you hold no citations, your signal is noisy. If you hold citations that never appear in any deal record, your instrumentation is leaking. Done when finance accepts the range as a standing line in the pipeline review.

That reconciliation only works if you know your citation position. Measuring citation share across LLMs gives you the input side and the CRM gives you the outcome side. When those two lines move together, you have about as much proof as this channel allows.

Where instrumentation stops and the engine starts

Instrumentation measures. It does not move anything. A perfectly instrumented CRM with no citations reports zero with excellent precision.

Phase four needs a citation record to check your AI Influenced deals against, and that record is the part OnlyAEO supplies through its LLM Attribution product. The monthly report gives you citation share, mention rate and LLM recall frequency. Put those next to your deal list: if Claude recall is rising and no deal evidence mentions Claude, the leak is in your forms or your call search, not in the channel.

First measurable citation improvements usually take at least 60 days, so a reconciliation run before then compares a flat citation line with a thin evidence log. Treat earlier reports as a test of the plumbing, and watch the AEO metrics that predict pipeline in the meantime.

If you do not know your starting position, the free AI visibility audit covers ChatGPT, Claude, Gemini and DeepSeek and arrives within 48 hours. It gives phase one a citation baseline to report against. The rest of how OnlyAEO works is the input side: content structured so engines can quote it, clear entities, and feeds machines can read.

Build the CRM instrumentation first anyway. It earns no citations. What it does is answer "what did it produce?" in the quarter budget gets reviewed, and in our experience the AEO programs that get cut are the ones that cannot.

Frequently asked questions

Does HubSpot track traffic from Claude automatically?

Yes, when the buyer clicks a link in Claude. HubSpot sets the traffic source to AI Referrals and records claude.ai as the platform in the first drill-down property. It cannot see buyers who read a Claude answer and visit later through search or by typing your URL, so those need a self-reported field and a deal-level flag.

How do I get AI-sourced deals into Salesforce campaign influence?

Create a campaign for AI answer engines, add leads and contacts as members when the referrer or the self-reported field names an assistant, and make sure the demo booker is a contact role on the opportunity. Campaign influence only credits campaigns whose members hold a contact role on the opportunity.

Should I use HubSpot's native AI Referrals source or build a custom property?

Use the native value as your base, because attribution reports group interactions by the session's traffic source and a custom field does not become one. Keep custom fields for what the native source cannot express, such as deal-level influence and the evidence behind it.

How do I stop reps from overwriting the traffic source on a contact?

Lock the property or restrict edit permissions. In HubSpot, manually changing a record's traffic source clears the drill-down properties, and that detail cannot be restored by hand because drill-downs are set automatically.

Is self-reported attribution reliable enough to report to a board?

Only as a range with the method attached. Buyers give vague answers, pick the first option in a list, and skip optional questions. Randomize the option order, make the field required, pair it with free text and call evidence, and report a low and high estimate rather than a single figure.