AI engines cite funded B2B startups only 10% of the time.
This is a row-level AEO/GEO benchmark of how often ChatGPT, Gemini, Claude, and Perplexity name funded B2B AI startups when buyers ask real category questions. We tested 30 companies across 5,400 engine responses — then published the raw dataset so journalists, founders, and AI crawlers can verify the finding.
For context: most published "AI search studies" test 1–5 prompts a single time and screenshot the result. Wave 1 of this study tested 15 buyer prompts 3 times each across 4 live AI engines — 5,400 measured responses, with 95% confidence intervals on every rate.
What we actually did
We typed 15 buyer questions into 4 AI search engines (ChatGPT, Gemini, Claude, Perplexity), asked each question 3 times to make sure the answer wasn't a fluke, and then counted — line by line — which companies got named.
10%
of answers named a funded startup
34%
named an established incumbent
15
real buyer questions tested
3×
we asked each question 3 times
Run 19b59b1a · May 31, 2026 · 4 AI engines · 5 B2B categories · raw data published free
The AI Citation Gap is the measurable difference between how often AI search engines name venture-funded B2B startups versus established incumbents when buyers ask category questions. In the May 31, 2026 run, four major AI engines were asked 15 buyer-intent prompts across 5 B2B AI categories — 5,400 total trials, n=3 runs per cell, with Wilson 95% confidence intervals on every reported rate. The funded-startup citation rate landed at 10% (95% CI 8–12%), versus 34% (95% CI 29–38%) for the incumbent control group on the same prompts. The engines clearly know the categories. They just don't know the funded vendors yet.
Key Takeaways
- Funded B2B AI startups were cited just 10% of the time (68/684 expected-match trials) when buyers asked the kind of category questions investors and prospects actually type into AI search.
- Incumbents got cited 34% of the time on identical prompts — proof that the engines understand the categories. The gap is who the engines know, not whether they know.
- The gap was not uniform across engines. ChatGPT and Perplexity cited funded startups noticeably more than Gemini and Claude, where most categories produced zero funded-vendor mentions.
- A handful of companies captured most of the funded-startup citations. Vendors with strong developer authority (Qdrant, Chroma) or category authority (Avoma) dominated; multiple well-funded competitors disappeared entirely.
- Funding does not buy AI visibility. Editorial coverage, structured data, a Wikidata entity, and category-authority content do. AEO and GEO are the path from invisible to cited.
Why we ran this study
Founders, marketers, and investors keep asking the same question: do AI engines actually name our vendor when buyers describe the problem? Anecdotes were everywhere; published, reproducible measurement was not. We built this study to answer the question with row-level, downloadable evidence — not vibes.
How deep this went
5,400 engine responses · 15 buyer prompts · 4 live AI engines · 30 companies (19 funded + 11 controls) · n=3 runs per cell · Wilson 95% CIs on every rate · sampling frame verified against public funding records · word-boundary + URL-host mention detection · full row-level CSV and JSON published under CC BY 4.0.
What it means for founders & marketers
If you raised a round and AI engines still don't name you when buyers describe your category, you have an AEO/GEO problem — not a product problem. The fix is third-party category authority, an entity-graph buildout, and structured data the engines can quote. The Authority Buildout playbook is exactly that, built from the patterns the cited vendors share.
Limitations & how to read this study
- Sample frame: 19 venture-funded AI startups + 11 category controls, funding dates verified against public records as of May 2026.
- Engines drift: ChatGPT, Gemini, Claude, and Perplexity update model weights frequently. Results are a snapshot of May 31, 2026, not a permanent ranking.
- Run count: n=3 runs per prompt × engine cell. Reported 95% Wilson CIs reflect this — narrow categories should be read with caution.
- Measurement: Mention detection uses word-boundary matching on company name + URL host on the engine's verbatim response. False positives/negatives are possible on short brand names.
- Prompt scope: Buyer-intent + investment-style prompts only. Brand-name lookups ("what does Avoma do?") are deliberately excluded — every engine handles those.
- Not a vendor ranking: Low citation does not mean low product quality. It means the AI search surface does not currently know the vendor for that buying conversation.
Wave 2 is in the field. The next release expands to ~50 buyer prompts and ~100 companies across the same engines, with re-tested Wave 1 prompts to measure model drift over time. Get the Wave 2 dataset on release →
Rows analyzed
5,400
row-level engine responses
Buyer prompts
15
searched across the engines
Companies
30
19 funded startups + 11 controls
Categories
5
investment-style AI markets
Funded startups vs. incumbent controls
The control group proves the engines know these categories. Funded startups are the visibility blind spot.
Key findings
- Overall funded-startup citation rate
- 10% (68/684 expected matches, 95% CI 8–12%)
- Overall incumbent citation rate
- 34% (133/396, 95% CI 29–38%)
- Citation gap (incumbent ÷ funded)
- 3.4× — incumbents cited that many times more often than funded startups on the same buyer prompts
- Engine most friendly to funded startups
- ChatGPT (GPT-5-mini) at 13%
- Engine least friendly to funded startups
- Gemini 3 Flash at 6%
- Worst category for funded startups
- Legal tech AI at 2% (incumbents: 31%)
- Prompt validity
- 29% of prompts surfaced at least one funded startup (53/180)
- Sample size
- 5,400 engine responses · 30 companies · 15 prompts · 4 engines
Comparison definitions
Exactly what we compared
This study is not comparing “small companies” to random big companies. It compares venture-funded B2B AI startups against established category controls in the same market, using the same buyer-intent prompts.
- Funded startup
- A venture-backed B2B AI company that has raised institutional funding — typically Seed through Series C (industry-standard ranges: roughly $2M–$150M total raised, valuations under ~$2B) — but is not yet the default category name an AI engine reflexively cites. In this study: Qdrant, Chroma, Avoma, Replicant, Regie.ai and the rest of the 19-company startup sample.
- Incumbent control
- An established category leader used as a control group to verify the engines understand the market — typically late-stage or category-defining vendors (industry-standard ranges: roughly $100M+ raised or $1B+ valuation, or dominant market share in the category). If the engine can name these companies, the prompt is valid. In this study: Pinecone, Glean, Gong, Harvey, Sierra, Cresta, Decagon and the rest of the 11-company control sample.
Why this matters: both groups were tested with the same prompts in the same engines. When incumbents are cited and funded startups are missing, the finding is an AI visibility gap — not proof that the category was unclear or that the engine failed the task.
By AI engine
- Funded startups
- Incumbent controls
By investment category
- Funded startups
- Incumbent controls
The citation gap, per engine
How many times more often each engine cites incumbents than funded startups on the same buyer prompts.
How to read this: a 7.6× bar means that engine named an established incumbent 7.6 times more often than a funded startup when asked the same buyer question. Higher = bigger visibility gap to close.
Bar scaled against a 10× ceiling, sorted worst-to-best. "∞" means the funded-startup rate rounded to 0%.
Who actually got cited
Citation was not evenly distributed. A few companies with existing developer or category authority captured most funded-startup mentions; several venture-backed companies disappeared entirely.
| Company | Category | Cited | Rate |
|---|---|---|---|
| Qdrant | Vector DB / AI infrastructure | 30/36 | 83% |
| Chroma | Vector DB / AI infrastructure | 15/36 | 42% |
| Avoma | Sales AI / RevOps | 9/36 | 25% |
| Replicant | Customer-service AI agents | 5/36 | 14% |
| Regie.ai | Sales AI / RevOps | 4/36 | 11% |
| Vectara | Enterprise knowledge / RAG | 3/36 | 8% |
| Robin AI | Legal tech AI | 2/36 | 6% |
| Crescendo | Customer-service AI agents | 0/36 | 0% |
| Dust | Enterprise knowledge / RAG | 0/36 | 0% |
| Garden | Legal tech AI | 0/36 | 0% |
| LanceDB | Vector DB / AI infrastructure | 0/36 | 0% |
| LegalFly | Legal tech AI | 0/36 | 0% |
| Lindy | Customer-service AI agents | 0/36 | 0% |
| Lorikeet | Customer-service AI agents | 0/36 | 0% |
| Marqo | Vector DB / AI infrastructure | 0/36 | 0% |
| Mem | Enterprise knowledge / RAG | 0/36 | 0% |
| Nooks | Sales AI / RevOps | 0/36 | 0% |
| Rilla | Sales AI / RevOps | 0/36 | 0% |
| Substrata | Sales AI / RevOps | 0/36 | 0% |
Writing about this study?
Republish under CC BY 4.0 with credit to Smart Money Media. Citations, raw data, and hi-res charts are below.
Citations pinned to Wave 1 — May 31, 2026, 10% (68/684, 95% CI 8–12%).
What we searched for
These were investment-style and buyer-intent category prompts, not brand-name prompts.
What are the leading vector database platforms for production AI?
Funded-startup mentions: 20/48 across 4 engines
Best vector databases for RAG and semantic search at scale?
Funded-startup mentions: 15/48 across 4 engines
Top vector search infrastructure for enterprise AI applications?
Funded-startup mentions: 10/48 across 4 engines
What are the leading AI sales coaching and conversation intelligence platforms?
Funded-startup mentions: 8/60 across 4 engines
Which AI voice agents are leading for customer support automation?
Funded-startup mentions: 4/48 across 4 engines
Best AI tools for sales prospecting and outbound automation?
Funded-startup mentions: 4/60 across 4 engines
Top RAG-based enterprise assistants buyers should evaluate?
Funded-startup mentions: 3/36 across 4 engines
What are the best AI contract review platforms for law firms?
Funded-startup mentions: 2/36 across 4 engines
Top AI agents for autonomous customer ticket resolution?
Funded-startup mentions: 1/48 across 4 engines
Top RevOps and revenue intelligence platforms buyers should evaluate?
Funded-startup mentions: 1/60 across 4 engines
Detailed results
Rates below use only expected category matches: the company's own market, the engine queried, and the prompt category.
Engine citation rates
95% CI 3–10%
95% CI 5–14%
95% CI 8–18%
95% CI 9–19%
Category citation rates
95% CI 24–39%
95% CI 4–12%
95% CI 1–8%
95% CI 1–8%
95% CI 1–7%
›Full engine × category citation matrix(funded targets only · % cited, with n)
| Engine | Sales AI / RevOps | Customer-service AI agents | Enterprise knowledge / RAG | Legal tech AI | Vector DB / AI infrastructure |
|---|---|---|---|---|---|
| ChatGPT | 7% 3/45 | 3% 1/36 | 0% 0/27 · wide CI | 0% 0/27 · wide CI | 50% 18/36 |
| Gemini | 4% 2/45 | 0% 0/36 | 0% 0/27 · wide CI | 0% 0/27 · wide CI | 22% 8/36 |
| Perplexity | 7% 3/45 | 3% 1/36 | 11% 3/27 · wide CI | 7% 2/27 · wide CI | 17% 6/36 |
| Claude | 11% 5/45 | 8% 3/36 | 0% 0/27 · wide CI | 0% 0/27 · wide CI | 36% 13/36 |
Read across each row to see how one engine performs across markets, or down each column to see which engine cites funded startups most in a single market. Cells with n < 36 carry wide confidence intervals and should be read as directional. Darker shading = higher citation rate.
When the engines did cite
Verbatim evidence from the test runs.
“Avoma / ExecVision (meeting-focused) — lower-cost alternatives that emphasize meeting intelligence, coaching and summaries for sales enablement.”
Avoma · ChatGPT (GPT-5-mini) · Sales AI / RevOps
“**If you prioritize sales coaching and call insights:** Choose **Gong**.”
Gong · Gemini 3 Flash · Sales AI / RevOps
“If you’re evaluating **RevOps** and **revenue intelligence** platforms, the most common buyer shortlist should include **Salesforce Sales Cloud, HubSpot, Gong, Clari, Revenue.”
Gong · Perplexity Sonar · Sales AI / RevOps
“Harvey — an AI assistant focused on legal research, document Q&A, and drafting; positioned toward law-firm workflows and ethical/compliance controls.”
Harvey · ChatGPT (GPT-5-mini) · Legal tech AI
“Top law firms most commonly use **professional-grade legal AI research platforms** rather than general-purpose chatbots, with the strongest recurring names in the sources being **CoCounsel**, **Harvey**, **Lexis+ AI**, and **Westlaw/Westlaw Precision**.”
Harvey · Perplexity Sonar · Legal tech AI
“Chorus — Conversation intelligence (direct competitor to Gong) focused on deal coaching and rep enablement.”
Gong · ChatGPT (GPT-5-mini) · Sales AI / RevOps
Methodology
The study was designed to measure AI search visibility, not traditional search rankings. Every number on this page is generated from row-level engine output stored in the dataset.
- Engines tested
- ChatGPT (GPT-5-mini), Gemini 3 Flash, Claude Sonnet 4.5, and Perplexity Sonar.
- Company frame
- 19 funded AI startups compared with 11 established category controls.
- Prompt frame
- 15 buyer-intent and investment-style prompts across Vector DB / AI Infrastructure, Sales AI / RevOps, Enterprise Knowledge / RAG, Customer-service AI Agents, and Legal Tech AI.
- Headline metric
- Citation rate = mentions ÷ expected-match trials, restricted to the company's own category.
- Control check
- Incumbent health = established vendors cited on their own category prompts. This prevents treating a silent category as a startup-specific failure.
- Run date
- May 31, 2026
| Total engine responses | 5,400 |
|---|---|
| Runs per company × engine × prompt | 3 |
| Engines tested | 4 |
| Buyer-intent prompts | 15 |
| Investment categories | 5 |
| Funded startups in frame | 19 |
| Incumbent controls in frame | 11 |
| Measurement | Word-boundary regex on engine output |
| Confidence intervals | Wilson score 95% |
| Dataset license | CC BY 4.0 |
| Run date | May 31, 2026 |
| Batch ID | 19b59b1a |
Cite this study
Published under Creative Commons Attribution 4.0. Reporters, analysts, and academics are welcome to republish with attribution.
Smart Money Media Team. (2026, May 31). AI engines cite funded B2B startups only 10% of the time. Smart Money Media. https://smartmoneymedia.org/research/ai-citation-gap
"AI Engines Cite Funded B2B Startups Only 10% of the Time." Smart Money Media, 31 May 2026, smartmoneymedia.org/research/ai-citation-gap.
@techreport{smartmoneymedia2026aicitationgap,
title = {AI Engines Cite Funded B2B Startups Only 10\% of the Time},
author = {{Smart Money Media Team}},
year = {2026},
month = {May},
institution = {Smart Money Media},
url = {https://smartmoneymedia.org/research/ai-citation-gap},
note = {CC BY 4.0. Row-level dataset: https://frdenvfzmvxhurwaxpcb.supabase.co/functions/v1/study-dataset?format=csv}
}Press & republication
- License: CC BY 4.0 — republish the charts and tables with credit to Smart Money Media and a link back to this page.
- Hi-res charts: Available on request for editorial use.
- Interview requests: The team is available for journalist comment on AI search visibility, AEO, GEO, and B2B AI marketing trends.
- Raw data: The full row-level CSV and JSON datasets are downloadable above. Methodology is fully documented.
- Contact: contact@smartmoneymedia.org
Frequently asked questions
What buyers, founders, and journalists ask about the study.
What is the AI Citation Gap?
The AI Citation Gap is the measurable difference between how often AI search engines name venture-funded B2B startups versus established incumbents when buyers ask category questions. In this study, funded startups were cited in 10% of expected-match trials, while incumbents were cited in 34% on the same prompts.
How was the AI Citation Gap study conducted?
We ran 5,400 AI search trials across 15 buyer-intent prompts, 5 B2B AI categories, and 4 engines (ChatGPT GPT-5-mini, Gemini 3 Flash, Claude Sonnet 4.5, Perplexity Sonar). Every prompt × engine cell was repeated n=3 times and reported with Wilson 95% confidence intervals. Mention detection used word-boundary matching on company name plus URL host. The complete row-level dataset is published under CC BY 4.0.
Is this a ranking of which AI startups are best?
No. Low citation does not mean low product quality — it means the AI search surface does not currently know the vendor for that buying conversation. This study measures visibility inside AI answers, not product fit, traction, or revenue.
Why do incumbents get cited more than funded startups?
Incumbents have longer-lived tier-1 editorial coverage, broader Wikipedia and Wikidata footprints, and denser long-tail citations on the open web. AI engines reflect those signals. Funded startups often have strong investor PR but thin third-party category coverage — exactly what AEO and GEO programs are designed to fix.
How can a startup get cited more often by ChatGPT or Perplexity?
Earn third-party editorial coverage on tier-1 outlets for the category, not just funding announcements. Publish original research the engines can quote. Ship structured data (Organization, Product, FAQPage), a Wikidata entity with verified sameAs links, and an llms.txt. Most importantly, control the entity graph so engines have a consistent answer to 'what is this company.'
Can I republish the charts and data?
Yes — under Creative Commons Attribution 4.0 (CC BY 4.0). Credit Smart Money Media and link back to this page. Email contact@smartmoneymedia.org for high-resolution chart exports or interviews.
How often will this study be updated?
AI engines update model weights frequently, so the current results are a snapshot of May 30, 2026. We plan to re-run the full study quarterly and publish a delta versus the previous batch. Historical batches remain accessible via the dataset endpoint.
See whether AI engines cite you.
Run our free AI Visibility Audit. We test your brand against the same engines used in this study and show exactly where you stand.