Your buyers shortlist in ChatGPT. How do you prove it shows up in pipeline?
AI-influenced buyers often arrive as direct or branded traffic, so referral reports undercount them. A measurement plan a finance team will accept.
The short answer
Not from referral traffic alone, because buyers who shortlist in ChatGPT can arrive later by typing your name. Pair a weekly share-of-answer series with branded search, direct traffic, an AI-assistant option in your how-did-you-hear field and AI referrals. Write down what you expect before the work starts, and read the trend over a quarter.
Key takeaways
- In G2's March 2026 survey of 1,076 B2B decision makers in North America, EMEA and APAC, 51% said they start software research with an AI chatbot more often than with Google.
- Referral reports undercount AI influence, because a buyer who shortlists in a chatbot can arrive later by typing your name.
- Five signals read together show AI influence: weekly share of answer, branded search, direct visits, self-reported source and AI referrals.
- Write down the expected movement before the work starts, so the read-out cannot be fitted to the data after the fact.
- In TrustRadius's January 2026 survey of 1,862 technology buyers, 83% shortlisted three or fewer products, so the first list is where the decision narrows.
In this article
You prove it by reading several signals together over a quarter, not by the AI referral line on its own. A buyer who builds a shortlist in ChatGPT often visits later by typing your name, so the visit lands in direct traffic or branded search. This guide from Sigzen AI sets out five signals, the analytics and CRM fields that capture them, and a 90-day plan a finance team can check.
Where do B2B buyers start now?
Many start in a chatbot. In G2's March 2026 survey of 1,076 B2B decision makers across North America, EMEA and APAC, 51% said they start software research with an AI chatbot more often than with Google1. TrustRadius found a lower but still large share in its January 2026 survey of 1,862 technology buyers from its global network: 63% used AI at some point in their buying process2.
The shortlist forms early and stays short. In the same TrustRadius survey, 83% of buyers shortlisted three or fewer products2. G2's respondents also said the chatbot changed the outcome: 69% reported that AI chatbots surfaced information that led them to choose a different vendor than expected3.
Our view: for a B2B SaaS company, the open question is no longer whether buyers use AI answers. It is whether your brand is on the list those answers produce, and whether you can show that being on it reaches pipeline. The rest of this piece is about the second half.
B2B software buyers, 2026
The shortlist often starts in a chatbot
Why AI influence hides in your analytics
It hides because most of it never arrives as an AI referral. Across 101,574 websites in 250 countries, SE Ranking found that AI search sent 0.32% of all visits in January–April 2026, up from 0.02% in 20244. SE Ranking sells search tools, so read this as a vendor study with a large sample. ChatGPT sent 74.78% of those AI referrals in the same months, down from 79.74% in 2025, with Gemini at 11.56%5.
A small referral line is real, but it measures one path. A buyer who reads a shortlist in an answer, closes the tab and searches for your name a week later shows up as branded organic search. One who types your address shows up as direct. Neither visit carries any trace of the chatbot.
What your analytics can already see
Google Analytics 4 now has an AI Assistant channel in its default channel group, which Google's help page describes as visits from sources like ChatGPT, Gemini, DeepSeek, Copilot or Grok6. That catches the click-through path, when the referrer survives the trip.
Themeisle's guide to tracking AI referrals in Google Analytics covers referrer filters, a custom channel group and landing-page analysis, and discusses why referrals are undercounted7. It is a good place to start. This piece picks up where the click is lost: four more signals, an expectation written down in advance, and a quarterly read-out.
How an AI shortlist reaches your CRM
- Buyer asks an AI engine for a shortlist
- Your brand is named, or not
- Later visit by name: branded search or direct
- Demo or trial form
- Opportunity in the CRM
What changes when you are cited?
Being cited changes behaviour on the results page, and that part you can partly measure. In Seer Interactive's panel of 53 brands, tracked from January 2025 to February 2026, pages cited in a Google AI Overview earned 120% more organic clicks per impression than when they were not cited8. Seer is an agency that sells search services, and the study covers Google's AI Overviews, not chatbots.
The same data carries a warning. Cited pages still earned fewer clicks per impression than queries that showed no AI Overview at all8. Citation softens a loss rather than adding traffic, so the clicks you keep are a weak proxy for influence.
Our view: treat citation as the leading indicator and pipeline as the lagging one. A rise in share of answer that is not followed, a few weeks later, by movement in branded search, direct visits or what buyers tell you is a reason to re-check the prompt set. It is not a reason to declare success.
Five signals that together show AI influence
No single signal proves influence, but five moving together make a case. Each one has a known bias, so the method is to read them side by side and to know which way each one leans.
| Signal | What it measures | Its bias | How to collect it |
|---|---|---|---|
| Share of answer, weekly | How often engines name you on buying prompts | Sampling noise; depends on the prompt set | Fixed prompts, repeated runs, same engines and countries |
| Branded search | Demand for your name in Google | Also moves with ads, press and events | Search Console branded query filter, weekly clicks |
| Direct visits | Visits by typed address or saved link | Includes staff, bots and lost referrers | Direct channel on key pages, internal traffic excluded |
| Self-reported source | What the buyer says sent them | Memory and wording | Free-text "how did you hear" field plus an AI option |
| AI referrals | Click-throughs that keep a referrer | Undercounts; app visits often lose it | AI Assistant channel or a custom channel group |
Our summary of each signal's bias. Channel and filter names from Google's help pages.69
Share of answer, weekly
Share of answer is the proportion of AI answers that name your brand across a fixed set of buying prompts. Run the same prompts on the same engines, in the same countries, every week. Repeat each prompt several times, because answers vary from run to run, and report a margin of error with every number.
For a SaaS company, the prompts are the ones a buyer types before a demo: "best help desk software for a growing support team", "alternatives to" a named rival, "X vs Y". Our measurement spec for AI visibility scores covers how many runs that takes and what to ask any vendor for, including us. The weekly series matters more than any single reading, because the other four signals are compared against its turning points.
Branded search and direct visits
Branded search counts people who already know your name. Google Search Console's Performance report has a branded and non-branded query filter, with data from March 2025. Google notes that some queries may be misclassified and that sites with few impressions do not get the filter9. Track weekly branded clicks and impressions, by country if you sell in several.
Direct visits need narrowing before they are useful. Watch direct sessions that land on pricing, comparison, integration and demo pages, with internal traffic excluded, rather than total direct traffic. Compare each week with the same week a year earlier where you can, because both series are seasonal. Both signals also respond to paid campaigns, events and press, so log those dates next to the series.
What buyers tell you
The cheapest signal is to ask. Add a required free-text field, "How did you first hear about us?", to demo, trial and contact forms. Next to it, offer an optional pick-list that includes "AI assistant (ChatGPT, Gemini, Perplexity, Copilot, Claude)".
Keep the raw text. Buyers write "ChatGPT recommended you" or "saw you in a comparison", and that wording is evidence a dropdown would flatten. Code the text weekly into a few buckets, with one person doing the coding so the rules stay stable. Ask the same question on the first sales call and record the answer, because some buyers only remember the chatbot once they are talking. Self-reported data is imperfect memory, yet it is the one signal that names the cause in the buyer's own words.
AI referrals and their landing pages
AI referrals are the click-throughs that keep a referrer, and the landing page tells you which answer sent them. In Google Analytics 4, start with the AI Assistant channel in the Traffic acquisition report6.
If you need engines it does not list, build a custom channel group. Google's help page explains that a channel can match the session source by regular expression, that traffic falls into the first channel it matches, and that custom groups can be applied to past data10. Place your AI channel above Referral so those sessions are not swallowed. Then pair the channel with landing page and key events such as demo requests. A cluster of AI visits on one comparison page often points to the answer that names you.
How to connect them without fooling yourself
Connect the signals with fields captured at the moment of conversion, and with an expectation written before the work starts. Without both, any rise in any series can be read as success.
Put the evidence in the CRM
When a form is submitted, pass hidden fields with the session's source, medium and landing page into the CRM record. Store the self-reported answer as two fields: the raw text and the coded bucket. Add an "AI touch" field with three values, yes, no and unknown, set from either the channel or the buyer's words.
Copy all of these onto the opportunity when it is created, so they survive a contact merge. A pipeline report can then split opportunities, stages and value by AI touch, which is a view a revenue team already trusts.
Write the expectation down first
Before the work starts, write one page. Say which prompts you expect to move and by roughly how much, which signals should follow, and how many weeks the lag should be. Date it and share it with finance, then compare the read-out against that page and nothing else.
This is pre-registration, borrowed from clinical research. It stops a team from finding a pattern after the fact in whichever series happened to rise. Our view: a modest prediction that comes true is worth more to a finance lead than a large uplift discovered later.
What does a quarterly read-out look like?
It looks like a short table that compares each signal with what was written on day 0, followed by a plain verdict. The example below shows the shape.
| Signal | Written on day 0 | Seen on day 90 | Read |
|---|---|---|---|
| Share of answer, 150 prompts | From about 1 in 10 answers to 1 in 7 | About 1 in 6 | Ahead of plan |
| Branded search clicks | Up on the same weeks last year | Up in two markets, flat in one | Mixed |
| Direct visits, key pages | Up on pricing and demo pages | Up, mostly on pricing | Consistent |
| Self-reported AI assistant | At least 10 demo requests | 14 of 120 demo requests | Consistent |
| AI Assistant channel | Up from a small base | From 200 to 350 sessions a quarter | Consistent |
| Opportunities with an AI touch | Tracked, no target set | 9, of which 3 in late stage | Too early for revenue |
Illustrative. Fictional company; invented, rounded numbers.
Northwind's read-out said four of the five pre-registered signals moved as written, one market did not, and revenue was too early to judge. It did not claim a return figure. Our view: that is the honest shape of a first-quarter result, and finance will trust the second quarter more because the first one admitted what it could not show.
What will a finance team accept?
Finance will accept a method it can audit: an expectation set in advance, several independent signals and honest limits. Finance is also in the room more often. G2 reports that finance involvement in software decisions jumped from 31% to 46% in a single year, in its Buyer Behavior Report published July 2026; the data period and region are not stated11. Finance will also ask what the work costs, and our dated list of published GEO prices sets agency and tool prices side by side.
Who else is in the decision
Buying groups are large. Forrester's survey of nearly 18,000 business buyers worldwide, fielded in 2025 (month not stated), found that on average 13 internal stakeholders and nine external participants influence a buying decision12. In the same survey (fieldwork month not stated), 94% of buyers reported using AI during their buying process13. Most of those people never fill in your form, so the self-reported field captures one voice among many.
The cost of missing the list is invisible in a pipeline report. If 83% of technology buyers shortlist three or fewer products, as TrustRadius found in January 2026, a vendor left off the list rarely sees the deal at all2. Your pipeline report only counts the deals you were in.
Why the AI-use figures disagree
Published shares of buyers who use AI range widely, and that is expected. In an older Gartner survey of 646 B2B buyers fielded in August–September 2025 (geography not stated), 45% said they had used AI during a recent purchase; those data are now more than a year old14. The three figures differ in sample, question and date.
Forrester's 94% (fieldwork month not stated) asked business buyers about their buying process13. TrustRadius's 63% asked technology buyers in January 2026 whether they used AI at any point2. Gartner's 45% asked about one recent purchase, a year earlier14. Our view: quote the figure whose question matches your claim, and never average them.
Share of buyers who used AI, by survey
A 90-day measurement plan
Ninety days is long enough for a shortlist change to reach first sales calls, and short enough to hold a budget holder's attention. The plan below is illustrative: the weeks are our planning assumptions, not measured lags, and a long sales cycle needs a second quarter before revenue shows.
Give each part an owner. Marketing owns the prompt set and the weekly series. Revenue operations owns the form fields and the CRM copy rules. Finance co-signs the expectation page on day 0 and chairs the read-out on day 90, which is what makes the result credible to the people who approve the next budget.
Ninety days from baseline to read-out
Prompt set frozen, first share-of-answer reading, AI channel and form fields live, expectations written and signed.
Plumbing check: hidden fields reach the CRM and copy to opportunities; free-text answers coded.
First comparison of share of answer with branded search and direct visits; campaigns and events logged.
Read-out against the day-0 page, signal by signal; next quarter's prompts and fixes agreed.
What this cannot tell you
This method shows whether AI influence is consistent with your pipeline data. It does not prove cause. We have not found a published primary study of how AI-referred visits convert in B2B software; the conversion figures that circulate come from US retail data, which says little about a long SaaS sale.
Buyers also check what chatbots tell them. In a Gartner survey of 645 B2B buyers fielded in August–September 2025 (geography not stated), 69% said they turn to sales reps to validate AI-generated insights; those data are now more than a year old15. So the AI touch is often the start of a deal that a person finishes, and a rep who hears "ChatGPT suggested you" should log it. If what the buyer read was wrong, our guide to correcting wrong AI answers covers the fix.
Our view, and we sell this work, so weigh it accordingly: if a quarter of honest measurement shows nothing moving, change the prompts or the content before you change the measurement. Start this week with the two cheapest steps, the how-did-you-hear field and the AI channel.
Sources
- G2, 2026 AI Search Insight Report: 1,076 B2B decision makers, North America, EMEA and APAC, Mar 2026
- TrustRadius, Beyond the hype: 1,862 technology buyers, Jan 2026 (Jul 2026); 94% is of buyers who used AI
- G2, 2026 AI Search Insight Report: 1,076 B2B decision makers, Mar 2026
- SE Ranking, AI traffic study: 101,574 websites, 250 countries, Jan–Apr 2026 (Jun 2026)
- SE Ranking, AI traffic study: share of AI referrals by engine, Jan–Apr 2026 (Jun 2026)
- Google, Analytics Help: default channel group, including the AI Assistant channel (checked Oct 2026)
- Themeisle, How to track AI referral traffic in GA4 (Sep 2026)
- Seer Interactive, AI Overviews and CTR, 2026 update: 53 brands, Jan 2025–Feb 2026 (Apr 2026)
- Google, Search Console Help: Performance report dimensions and the branded queries filter (checked Oct 2026)
- Google, Analytics Help: custom channel groups (checked Oct 2026)
- G2, Buyer Behavior Report (Jul 2026); data period and region not stated
- Forrester, State of Business Buying 2026: buying groups, fielded 2025 (Jan 2026)
- Forrester, State of Business Buying 2026: nearly 18,000 buyers worldwide, fielded 2025 (Jan 2026)
- Gartner, survey of 646 B2B buyers, Aug–Sep 2025; geography not stated (Mar 2026)(dated)
- Gartner, survey of 645 B2B buyers, Aug–Sep 2025; geography not stated (May 2026)(dated)
