When LPs ask AI for venture capital fund due diligence, who gets named?
We put 100 real limited-partner questions to ChatGPT and Gemini and Perplexity and checked which of 23 venture capital funds each answer named. 64 of the 100 questions got no named fund at all — this is the LP side of venture capital: fund selection, fees, due diligence and portfolio construction, never the founder side. Run the check for your firm →
Hover or tap a dot for the question.
Why we ran this
Choosing a venture capital fund is one of the least transparent allocation decisions an LP makes — track records are self-reported, benchmarks are inconsistent, and most due-diligence knowledge lives in closed networks rather than public content. Increasingly, a first pass at that knowledge starts with a question to an AI assistant, not a call to a placement agent. The assistant assembles its answer from whatever public content exists — fund sites, LP-education content, industry press, forums — and either names specific funds or it doesn’t. This benchmark reads how visible venture capital funds currently are on the LP-education side of the category, engine by engine, and where the content gaps are.
How we measured it
We wrote 100 questions in the language an institutional LP, family office allocator or fund-of-funds analyst actually uses — fund selection, fees and economics, due diligence and red flags, emerging vs. established managers, co-investment and LP reporting, and portfolio construction (all 100 are explorable in full below). Each question was sampled once per engine, and every answer was checked for a mention of any of 23 identifiable venture capital funds — from multi-stage crossover platforms like Sequoia Capital and Andreessen Horowitz to traditional partnership-model firms like Union Square Ventures and Greylock Partners — using the same name matcher a live Inserviss scan uses.
One well-known fund, Benchmark, is deliberately not tracked: its name collides with the generic finance term “benchmark” (as in “benchmark a fund against its peers”), and several of this study’s own questions use that exact word. A spot-check of the raw answers found most whole-word matches were the common word, not the firm, so it was excluded rather than publish an inflated count.
A single pass per engine establishes direction and priority; it is not a trend line. Citations are reported as measured; query volumes are labelled estimates, shown as ranges. Engines are never blended. This run covers ChatGPT and Gemini and Perplexity; Microsoft Copilot is not currently sampled.
This is deliberately the LP due-diligence angle, not the much larger founder-fundraising audience — every question is phrased as an allocator deciding whether and how to commit capital to a fund, never as a founder asking how to raise from one.
What we found, at a glance
- Visibility varies sharply by engine. ChatGPT named a tracked fund in 13 of 100 answers, Gemini named a tracked fund in 34 of 100 answers and Perplexity named a tracked fund in 15 of 100 answers.
- 64 of 100 questions got no tracked fund on any engine. Weakest: Fund Selection & Track Record, Fees, Terms & Fund Economics, Due Diligence Process & Red Flags.
- Sequoia Capital is the most-named fund, appearing in 41 of 300 answers. Only 21 of the 23 tracked funds were named even once.
- Multi-stage platforms and traditional venture firms get named at very different rates. See the full breakdown, and which archetype AI actually favors, in the table below.
How often does AI name a fund at all?
Below is the share of the 100 answers that named at least one tracked fund, per engine — then the full table, every fund against every engine, sortable by any column.
Share of 100 answers naming at least one of the 23 tracked funds. Per engine, never blended.
| Fund | ChatGPT | Gemini | Perplexity |
|---|---|---|---|
| Sequoia Capitalmulti-stage | 10 | 24 | 7 |
| Andreessen Horowitzmulti-stage | 11 | 12 | 9 |
| Founders Fund | 3 | 19 | 5 |
| Accelmulti-stage | 8 | 8 | 6 |
| Lightspeed Venture Partnersmulti-stage | 7 | 10 | 4 |
| Bessemer Venture Partnersmulti-stage | 5 | 9 | 5 |
| General Catalystmulti-stage | 4 | 6 | 5 |
| Index Ventures | 6 | 4 | 3 |
| Greylock Partners | 6 | 4 | 1 |
| Kleiner Perkins | 4 | 2 | 3 |
| Khosla Ventures | 4 | 3 | 2 |
| First Round Capital | 1 | 7 | 0 |
| Union Square Ventures | 1 | 5 | 2 |
| New Enterprise Associates | 3 | 1 | 3 |
| Emergence Capital | 0 | 6 | 0 |
| CRV | 3 | 1 | 1 |
| IVPmulti-stage | 1 | 2 | 1 |
| Insight Partnersmulti-stage | 0 | 1 | 2 |
| Tiger Globalmulti-stage | 1 | 1 | 1 |
| GV | 0 | 1 | 1 |
| Redpoint Ventures | 0 | 1 | 0 |
| Thrive Capitalmulti-stage | 0 | 0 | 0 |
| Foundation Capital | 0 | 0 | 0 |
Mentions across 100 answers per engine. Select a column heading to re-sort. Scroll the table sideways on a narrow screen.
Download this chart as an image →
Where the openings are
Coverage varies by question type. Below is the best engine’s coverage for each cluster, out of ~17 questions.
Answers naming a fund on the strongest engine, per cluster (out of ~17).
How this is estimated
Estimate — a range, not a keyword-tool export. Illustrative estimate, not a keyword-tool export. Each cluster has ~16-17 seed LP questions; this is a narrow, institutional search category (LPs and allocators, not the much larger founder-fundraising audience) — long-tail intent phrases run ~20–90 monthly searches each globally, one or two head terms add ~250–800/mo, and an AI-assistant reformulation multiplier of ×2–4 accounts for the wider phrasing people use with an assistant. Ranges are rounded and widened. For a specific fund, the real number comes from the check, not this table.
| Cluster | Best-engine coverage | Est. questions / mo |
|---|---|---|
| ●Fund Selection & Track Record | 9/17 | 1,600–6,200 |
| ●Fees, Terms & Fund Economics | 2/17 | 1,100–4,300 |
| ●Due Diligence Process & Red Flags | 1/17 | 1,000–4,000 |
| Emerging Managers vs Established Funds | 12/17 | 900–3,600 |
| ●Co-Investment, Reporting & LP Rights | 1/16 | 750–3,000 |
| ●Fund Strategy & Portfolio Construction | 9/16 | 850–3,400 |
Volumes are an estimate, shown as a range. Illustrative estimate, not a keyword-tool export. Each cluster has ~16-17 seed LP questions; this is a narrow, institutional search category (LPs and allocators, not the much larger founder-fundraising audience) — long-tail intent phrases run ~20–90 monthly searches each globally, one or two head terms add ~250–800/mo, and an AI-assistant reformulation multiplier of ×2–4 accounts for the wider phrasing people use with an assistant. Ranges are rounded and widened. For a specific fund, the real number comes from the check, not this table.
The check runs this same 100-question universe against your fund’s name, per engine, and returns your AI Visibility Score plus the exact questions where a competitor — or no one — is being named.
Check your fund’s AI visibility →Explore the 100 questions
Every question, and exactly what each engine did with it — which funds it named, and which sources it pulled from. Filter by cluster, search for your own name, or show only the questions no fund has claimed.
The 64 questions no tracked fund owns on any engine, with the sources AI cites instead — a ready-made content brief.
One email. No list-selling. Unsubscribe any time.
Where each engine gets its answers
The domains each engine actually pulled from, per engine — never blended. ChatGPT grounds on institutional LP-education sources — Cambridge Associates, the NVCA, CB Insights, PitchBook and ILPA — the closest thing to an authoritative source list for this category. Gemini’s citations are the thinnest and most scattered of the three, a long tail of independent VC newsletters and blogs with no domain cited more than twice. Perplexity leans hardest on LinkedIn (58 citations, more than any other engine’s top domain) plus a cluster of independent VC-commentary sites (govclab.com, vcbeast.com, valueaddvc.com) — social and opinion content, not institutional sources.
Times a domain was cited across the 100 answers. Blue = a tracked fund’s own site.
About this study
- Single pass per engine. One sample per question per engine establishes direction; it does not average out run-to-run answer variance.
- Snapshot in time. Collected September 14, 2026. AI engines change their answers week to week; this is not a trend line.
- Conservative name matching. Short one-token fund names (Accel, GV, CRV…) require an exact whole-word match, so a slight undercount is possible — but it cannot explain a 64-of-100gap. One fund, Benchmark, was excluded entirely for the opposite reason: its name collides with the generic finance term “benchmark” (see “How we measured it” above).
- ChatGPT and Gemini and Perplexity this run. Microsoft Copilot is not currently sampled; the measurement is identical when it is.
How to cite this study
To request the underlying data or discuss a fund-specific analysis, contact igor@inserviss.app.
The same measurement, run against your fund’s name — per engine, question by question — with your score and the questions to claim first.
Run my free scan →