GAP 1
Zero click
AI answers the buyer’s landscape question in full, so fewer buyers visit vendor sites.
68%
of US Google searches ended without a click, Jan–Apr 2026 (SparkToro)
(x,y) · RESEARCH · NOVEMBER 2026
The AI Search Zero Pick Report
1 in 5 B2B software vendors that AI names in UK-based searches about a market are never recommended when AI is given a real buying task.
The report explained · 3 min 42 · Nick Clark, founder
00 · Headline findings
Named isn't chosen.
We asked ChatGPT, Google AI Mode and Perplexity 90 questions across 15 B2B software categories, as a UK company with about 200 staff, and tagged 2,250 answers. Most of the vendors are global companies, but every search and question in the study was UK-based. Zero pick is the newest gap in AI search, and the most dangerous, because from the outside it looks like healthy visibility.
1 in 5
vendors risk zero pick. Of the vendors an engine names in its landscape answer, 20% are never recommended by that engine in any buying question (81 of 412).
17%
are never recommended by any of the three engines.
12%
of answers to buying questions got a single top pick. No question asked for one name.
62–76%
of recommendations go to the top three vendors in a category.
31%
The good news: of the vendors AI recommends in specific buying questions, 31% were missing from its own landscape answer.
01 · How we tested
Two stages, the way buyers use AI.
Each category was tested in two stages: first a question about the market in general, then a real buying task. Every vendor in the study lands in one of four groups in the AI answers.
| Stage | Group | What it means |
|---|---|---|
| 1. Market question | Named | Listed when we ask “What are the main options?” |
| 2. Buying task | Not put forward | Mentioned but not suggested for this buyer: background, an example only (“e.g. X”), ruled out or the wrong fit |
| 2. Buying task | Picked | Recommended to evaluate or buy: on a shortlist, in a comparison table, or “consider X” |
| 2. Buying task | Top pick | The one vendor the answer names as its first choice among the options it gives, with no “if you…” conditions |
Zero pick means named in stage 1 but never picked or selected as a top pick by the same engine in any stage-2 buying task.
02 · The three gaps
From zero click to zero pick.
AI search took the click; agents are now making the pick. Buyers meet three gaps on the way from a question to a shortlist.
GAP 2
Zero mention
AI doesn’t name you when buyers ask about your category. You are invisible, but at least you can see where.
Varies brand by brand
measure yours with a Datum Diagnostic
GAP 3 · THIS REPORT
Zero pick
AI names you, but never recommends you when it is given a real buying task. Your visibility looks healthy while you lose at the moment of decision.
1 in 5
named vendors never picked
AI search took the click; agents are now making the pick.
Zero click is well documented. 68% of US Google searches ended without a click in January–April 2026, up from 60% in 2024 (SparkToro, Similarweb data). When an AI summary appears, people click a result link on 8% of visits, against 15% without one (Pew Research Center, March 2025 data). This study takes the next step: what happens after the buyer knows the market and asks AI to help choose a vendor.
Agents are already in the buying group. Over 60% of B2B technology buyers use generative AI at some point in their evaluation (Forrester, The B2B Marketing Mix 2025). People still define the problem and sign the deal. Increasingly, agents do the research and the shortlisting in between, briefed with objectives rather than instructions. The tipping point is when the agent’s shortlist becomes the only shortlist.
Tools measure visibility. Datum Point measures whether you get picked.
03 · Finding 1
1 in 5 named vendors never get picked.
Across the three engines, 20% of the vendors AI names in its landscape answer are never recommended by that engine when asked a specific buying question. For example: “Our head of sales needs CRM software for a UK company with about 200 staff. Recommend what we should evaluate and why.”
A category is one software market, such as CRM, HR or accounting software; the study covers 15. These vendors are not invisible: the engine knows them and names them, but never puts them forward when the buyer is choosing. Each vendor had at least 20 chances: four buying questions per category, run daily for a week on ChatGPT and Google AI Mode, and five times each on Perplexity.
Perplexity 33 of 136 · ChatGPT 31 of 156 · Google AI Mode 17 of 120 engine–vendor pairs.
- It varies by engine. A vendor can be picked on one engine and invisible at the decision on another.
- It varies by market. Zero pick is highest in business intelligence (30%), identity and access management (29%), accounting (28%), ERP (27%) and HR (26%). It is lowest in marketing automation (4%) and contract management (9%).
- Some vendors miss everywhere. 30 of the 175 vendors named in landscape answers (17%) were never recommended by any engine in any buying question.
Who falls into the gap?
About half are vendors AI judged the wrong size for the buyer described: large enterprise suites, and tools built for very small businesses. Every question described a UK company with about 200 staff, so that part is a genuine fit judgement. The other half look like a reasonable fit for a 200-person UK company: mid-market suites, specialist tools, UK providers. Half of all zero picks were named in AI’s answer to “What are the main options for a UK company with about 200 staff?”, so AI itself listed them as options for this exact buyer, then never recommended them. That is the real gap: about 1 in 10 named vendors fit the buyer and still never get picked.
Buyer agents filter on fit before anything else. If AI can’t tell who you are right for, or thinks you are right for someone else, you are named in the market and left off the shortlist. Brand awareness alone doesn’t carry a vendor through.
04 · Finding 2
AI rarely commits, and a few vendors take most of the picks.
AI mostly hands buyers a shortlist, not a winner. The buying questions asked AI to recommend what to evaluate, advise which options fit, recommend what to move to, or prepare a CFO recommendation; none asked for one name. AI volunteered a single top pick in only 175 of 1,500 answers to buying questions (12%).
| Engine | Answers with one top pick | Top 3 vendors’ share of recommendations |
|---|---|---|
| ChatGPT | 21% (125 of 600) | 62% |
| Perplexity | 13% (40 of 300) | 76% |
| Google AI Mode | 2% (10 of 600) | 64% |
- Google AI Mode hedges by design. It sets out options by fit, then asks the buyer clarifying questions: industry, headcount, current systems. The shortlist is decided in that follow-up conversation.
- ChatGPT is the most decisive. It commits most often when the buyer describes a problem with an obvious next step, such as moving management reporting out of spreadsheets.
- Shortlists are concentrated. In most categories the same three names take around two-thirds of recommendations. Everyone else competes for what is left.
For a vendor, being on the shortlist is the realistic win. Making the shortlist depends on AI understanding exactly which buyer you are right for.
05 · Finding 3
The good news: specific questions open the door.
When buyers describe exactly what they need, AI looks beyond the usual names. 31% of the vendors AI recommended in buying questions (146 of 477) were not in its own landscape answer for that category. The more specific the question, the wider the door.
Problem question example: “Our contracts are stuck in email and we keep missing renewal dates.” Integration: “works with Shopify”, “works with Xero”.
The specialists that got in matched one precise need:
- a holiday and absence tracker, for a company still tracking leave in spreadsheets;
- month-end close tools, for a finance team whose close takes too long;
- a help desk built for ecommerce, for a Shopify business;
- a simple email marketing tool, for a three-person marketing team;
- a managed detection service, for a small IT team buried in alerts.
Google AI Mode does this most: 43% of the vendors it recommends were missing from its landscape answers. The door is open, not wide open. These vendors still win only 8% of all recommendations.
Vendors who are clearly the answer to one specific need can win the pick without winning the category.
This is good news for companies that could never outrank the big players in traditional search. Buyers increasingly ask AI very specific questions about problems, integrations and constraints. The work is making sure AI knows exactly which problem you solve, and for whom.
06 · Results by category
From 30% to 4%.
Zero pick ranges from 30% in business intelligence to 4% in marketing automation.
See the full table
| Category | Zero pick | Never picked by any engine | Top 3 share of picks | Answers with one top pick | Picks from outside the landscape |
|---|---|---|---|---|---|
| Business intelligence | 30% (6 of 20) | 4 of 11 | 73% | 39% | 59% |
| Identity and access management | 29% (9 of 31) | 2 of 12 | 79% | 12% | 12% |
| Accounting | 28% (7 of 25) | 4 of 11 | 65% | 6% | 28% |
| ERP | 27% (8 of 30) | 4 of 13 | 77% | 9% | 15% |
| HR | 26% (12 of 46) | 3 of 20 | 46% | 6% | 29% |
| Help desk | 24% (8 of 33) | 3 of 13 | 64% | 13% | 34% |
| CRM | 20% (4 of 20) | 2 of 10 | 76% | 11% | 36% |
| Cookie consent management | 20% (5 of 25) | 0 of 10 | 83% | 5% | 23% |
| Sales intelligence | 20% (5 of 25) | 0 of 10 | 73% | 7% | 9% |
| Endpoint protection | 14% (3 of 21) | 2 of 9 | 73% | 15% | 22% |
| Knowledge management | 14% (3 of 22) | 2 of 9 | 67% | 10% | 39% |
| Project management | 13% (3 of 23) | 1 of 9 | 67% | 17% | 20% |
| Expense management | 11% (4 of 35) | 2 of 15 | 57% | 2% | 26% |
| Contract management | 9% (3 of 32) | 1 of 14 | 56% | 11% | 47% |
| Marketing automation | 4% (1 of 24) | 0 of 9 | 65% | 12% | 28% |
| All categories | 20% (81 of 412) | 30 of 175 | 62–76% | 12% | 31% |
Zero pick counts are engine–vendor pairs across the three engines. Top 3 share is the average of the three engines. Category-level results only: vendor-level results are not published.
07 · What to do
How to avoid zero pick.
Zero click doesn’t have to mean zero pick. Six steps close the gap, each tied to what the study found.
01
Measure picks, not just visibility
Test the questions your buyers ask AI, by role and engine, and see whether you are recommended or only named.
From the studyVisibility tools would show every zero-pick vendor as visible.
Datum Diagnostic, then Assess: True Position Report
02
Make your fit unmistakable
State plainly who you are for: company size, sector, integrations, UK requirements.
From the studyAbout half of zero picks were judged the wrong size for the buyer described.
Build: content audit
03
Answer the question, not the keyword
Create content for the specific problems and integrations buyers describe.
From the study16% of picks on problem questions went to vendors outside the landscape answer.
Build: citable content
04
Win the sources agents trust
Earn credible third-party mentions: reviews, comparisons, trade media, communities.
From the studyAI checks what others say about you before it recommends you.
Build and Cited: earned media
05
Speak to every buyer, including the CFO
Give finance and procurement the cost, risk and value answers they ask AI for.
From the studyCFO questions were the least open: 3% of picks went to vendors outside the landscape answer.
Assess: buyer mapping
06
Track your pick rate
Watch recommendation, not just mention, across engines over time.
From the studyResults differ by engine: a pick on one can be a miss on another.
Cited: Navigator
What we won’t advise: quick GEO hacks. Google says AI search uses the same quality signals as search, and its spam policies now target attempts to manipulate AI visibility. The work that moves the pick is editorial and strategic first, then technical. You want to be found in AI search in the same way humans find you: authentic, natural, relevant communication.
This study did not audit every vendor in each market, so it does not measure zero mention. Datum Point can map the full market for any category, so you can see who is missing from the conversation altogether.
08 · The report
Get the full report.
The AI Search Zero Pick Report as a 10-page PDF: every finding, all 15 categories, the six steps and the method. Free to download and share.
YOUR REPORT
Thank you. It’s yours.
Your download should start now. If it doesn’t, download the PDF here. Want your own zero-pick rate? Get your free Datum Diagnostic.
09 · Method
Method and limitations.
The study mirrors how buyers use AI: first a question about the market, then a real buying task.
- Categories
- 15 B2B software categories, chosen from G2’s category structure.
- Questions
- Six per category, 90 in all, each framed for a UK company with about 200 staff. Two landscape questions (“What are the main X options?”, “Which X tools are most widely used?”) and four buying questions: a role-based question, an integration need, a problem to solve and a CFO recommendation with costs.
- Engines and dates
- ChatGPT and Google AI Mode answers collected through Peec AI from the UK, daily from 19 to 26 October 2026 (1,800 answers). Perplexity answers collected through its Sonar API on 19 October 2026, five runs per question (450 answers). 2,250 answers in total.
- Tagging
- Only vendors named in the answer text count; product names roll up to the parent company; integration partners and non-specialist tools (spreadsheets, file storage) are excluded. In a landscape answer every vendor counts as named. In an answer to a buying question each vendor is sorted into top pick, picked (recommended to evaluate or buy) or not put forward (background, an example only, ruled out or the wrong fit).
- Zero pick
- A vendor named in at least 20% of an engine’s landscape answers for a category, and never picked or selected as a top pick by that engine in any buying question in that category. Counting vendors given only as an example as picked barely changes the result. The named pool uses both landscape questions; building it from the 200-staff question alone gives 15% (about 1 in 7).
- Quality control
- Every answer was extracted and tagged against written rules. Every zero-pick vendor was then checked against the original answers: where it appeared in a buying task, the answer named it and then ruled it out, and no case was found of a missed recommendation.
Limitations
- One buyer profile. Every question described a UK company with about 200 staff, so some zero picks reflect genuine fit. Results for other company sizes and countries would differ.
- One week of answers. AI answers change from day to day and model to model; this is a snapshot, not a ranking.
- Perplexity was collected through its API, which can differ from the consumer app.
- We did not audit every vendor in each market, so the study does not measure zero mention.
- Zero click figures come from published research, not from this study.
- Category-level results only. Vendor-level results are not published.
Sources
- SparkToro, US Google zero-click searches January–April 2026, Similarweb clickstream data, as reported by TechWyse, 11 June 2026
- Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, 22 July 2025
- Forrester, The B2B Marketing Mix 2025, as quoted in Crackle PR, State of B2B Tech PR 2026
- Datum Point, AI Search Zero Pick study: 2,250 answers from ChatGPT, Google AI Mode and Perplexity, 19–26 October 2026
10 · Questions
About the study.
What is zero pick?
Zero pick is when AI names your company when buyers ask about your market, but never recommends you when it is given a real buying task. In this study, 1 in 5 vendors that AI named in its answer about a market were never recommended by the same engine in any buying question.
How is zero pick different from zero click?
Zero click is when AI answers a buyer's question in full, so they never visit your website. Zero pick comes later: the buyer knows the market and asks AI to help choose, and AI leaves you off the shortlist even though it knows you.
Which AI engines did you test?
ChatGPT and Google AI Mode, collected through Peec AI from the UK every day from 19 to 26 October 2026, and Perplexity through its Sonar API on 19 October 2026. 2,250 answers in total.
Is AI just filtering out vendors that don't fit?
Partly. Every question described a UK company with about 200 staff, and about half the zero picks were vendors AI judged the wrong size. The other half looked like a reasonable fit, and many were named in AI's own answer about the options for exactly that buyer.
Can I see my own company's result?
We don't publish vendor-level results. The free Datum Diagnostic tests the buying questions your buyers would ask AI and shows whether you are named, picked or missing.
Can I use the findings?
Yes. Please cite "Datum Point, The AI Search Zero Pick Report, November 2026" and link to datumpoint.co/zero-pick.
Start
Does AI pick you, or just name you?
Find out your zero-pick rate with a free Datum Diagnostic. Seven questions, about ten minutes. We run it within 48 hours and talk you through what we find.
Get your Datum Diagnostic