Company and individual details from this experiment have been generalized and anonymized throughout this article. The finding is about data availability, not the specific people or companies involved.
We asked ChatGPT to identify senior sales leaders at 10 B2B software companies, using public web research only. It identified a current, credible sales leader at 10 out of 10 companies.
Then we asked for what it actually takes to reach them.
0 out of 10 came back with a usable work email and direct phone number.
We ran the identical task with Lusha connected. 10 out of 10 records returned both.
The takeaway isn’t that AI is bad at prospecting. It’s surprisingly good at finding the right people. The gap is between finding someone and having what you need to actually reach them.
The experiment
We selected 10 real, active B2B software companies, ranging from early-stage to well-funded, across categories including developer tools, AI infrastructure, and enterprise software. Both conditions received the same fixed instruction:
“For each of the following companies, identify a current senior sales leader (VP Sales, SVP Sales, CRO, or equivalent). Return their full name, current title, work email, direct phone number, and LinkedIn profile. Do not guess or infer contact details. If a field cannot be verified, return ‘not found.'”
The ChatGPT-alone condition received only the company names and this instruction, public web research, no connected data source, no names supplied in advance.
The Lusha-connected condition used the same 10 companies, filtered through Lusha for US contacts, Sales department, VP or C-suite seniority, Software Development companies, B2B business model, 201-1,000 employees. That filter returned 2,161 matching contacts. We selected one qualifying sales leader at each of the 10 companies and revealed the same requested fields for all 10.
The scoring rubric was fixed before either result was reviewed:
| Metric | Scoring |
| Person identified | 1 if a qualifying leader is returned |
| Current role established | 1 if current title is confirmed |
| LinkedIn / profile | 1 if a person-specific professional profile is returned |
| Work email | 1 if a usable, non-masked address is returned |
| Direct phone | 1 if a usable, person-specific number is returned |
| Complete record | 1 if person + current role + work email + phone all present |
Where a company had more than one plausible senior sales leader, any current VP, SVP, or CRO-equivalent counted as a valid identification, we didn’t force an exact match to the specific person Lusha’s search happened to select.
The results
| Field | ChatGPT + public web | Lusha-connected |
| Current qualifying sales leader identified | 10/10 | 10/10 |
| Current role established | 10/10 | 10/10 |
| LinkedIn / profile information | 8/10 | 10/10 |
| Usable work email | 0/10 | 10/10 |
| Usable direct phone | 0/10 | 10/10 |
| Job start date | 0/10 | 10/10 |
| Complete, contactable record | 0/10 | 10/10 |
What ChatGPT actually did well
The first row is the one worth sitting with. ChatGPT didn’t fail at finding people, it identified a credible, current senior sales leader at every single company in the set, using nothing but public web research. In several cases it correctly surfaced role information from a company’s own announcements or leadership pages. Where multiple valid leaders existed at a company, it generally found a real one.
That’s a genuinely useful capability. It’s a real, current answer to “who runs sales at this company,” produced from an open-ended natural-language request instead of a filtered database search.
Where it fell apart: contactability
Under the “do not guess” rule, public search repeatedly surfaced partial or masked information, a partially obscured email pattern, a masked phone number, a “reveal contact” prompt pointing to a different data provider. None of those counted as a result, correctly, since the instruction was explicit: don’t guess, don’t infer, return “not found” if a field can’t be verified.
Under that honest standard, public web research returned zero usable work emails and zero usable direct phone numbers across all 10 companies. Not because the people don’t exist or aren’t findable in principle, but because a public, unstructured search doesn’t reliably surface verified, current contact information, even when it correctly identifies exactly the right person.
What we’re not claiming
Worth being precise about what this does and doesn’t show. This is a 10-company exploratory test, not a database accuracy benchmark. We measured whether the requested fields came back, not whether every email is deliverable or every phone number connects. We didn’t call the numbers or send the emails. And where a company had more than one valid senior sales leader, we scored any current, correct match, we didn’t require ChatGPT to name the exact same individual Lusha’s search happened to surface.
The honest, defensible finding is narrower than “Lusha is 100% accurate,” and it’s this: in this test, ChatGPT reliably identified the right kind of person, and reliably failed to produce a complete, usable contact record for them under a no-guessing standard. Connecting Lusha closed that specific gap, for these 10 companies, in this test.
AI is the reasoning layer. B2B data is the data layer.
This is the real finding underneath the numbers. ChatGPT’s research capability didn’t change between the two conditions, the interface, the reasoning, the ability to interpret “senior sales leader” and find a real candidate, all stayed constant. What changed was whether a structured, verified data source was available to answer the second half of the request.
Identifying the right person is a reasoning problem, and AI is already good at it. Producing a current, verified, contactable record is a data problem, and that’s a different job entirely.
Methodology
This experiment ran August 2026 against 10 real, active B2B software companies. Both conditions received an identical fixed prompt requesting a current senior sales leader’s name, title, work email, direct phone, and LinkedIn profile, with explicit instructions not to guess or infer any field.
The ChatGPT-alone condition used public web research only, with no connected data source and no contact names supplied in advance. The Lusha-connected condition filtered US contacts in the Sales department, VP or C-suite seniority, Software Development companies, B2B business model, 201-1,000 employees, a population of 2,161 matching contacts, from which one qualifying leader per company was selected and the same fields revealed.
Scoring followed a rubric fixed before either result was reviewed. Any current, correct VP, SVP, or CRO-equivalent leader counted as a valid identification, regardless of whether it matched the specific individual the Lusha search surfaced.
Company names and individual identities from this experiment have been generalized and anonymized throughout this article, including in aggregate discussion, since the pairing of a specific company with a specific named executive is itself identifying information. This is a 10-company exploratory test measuring field availability under a no-guessing standard, not deliverability, connectivity, or an independent benchmark of database-wide accuracy. Results from this sample should not be interpreted as representative of accuracy at scale.
FAQ
Can ChatGPT find B2B contacts?
It can reliably identify people who fit a role and company profile using public web research. In a controlled 10-company test, it identified a correct, current senior sales leader in 10 out of 10 cases. It did not reliably produce a complete, verified contact record, work email and direct phone, for any of them.
Can ChatGPT find business emails?
Under a strict no-guessing standard, public web research alone did not return a single usable work email across a 10-company test. Public sources often surface masked or partial contact information rather than verified, current details.
What’s the difference between identifying someone and being able to contact them?
Identifying someone is a reasoning task, matching a role and company to a real, current person, and general-purpose AI is genuinely capable at it. Contacting them requires verified, current data, a work email and direct phone that are actually correct, which is a structured data problem, not a research problem.
Why connect Lusha to ChatGPT?
Connecting Lusha gives ChatGPT access to verified, structured contact and company data while you keep working through a conversational interface, closing the gap between identifying a prospect and having a usable record to act on.
