A critical assessment of SuperMe’s $20M Series A led by Forerunner — a “network of AI expert profiles” that attaches real names to hallucination-prone output, contractually caps its liability at $100, and names no AI model vendor anywhere in its own legal documents.
The Terms do the work the marketing won’t: output “may contain hallucinations,” users “MUST evaluate the accuracy,” and liability is capped at $100. The product sells a named real expert’s authority while disclaiming all responsibility — and the reputational fallout lands on the real person whose name is on the profile.
A company whose entire pitch is AI names no LLM vendor — not on the site, not in the Terms, not in the Privacy Notice. Only Stripe and Google appear. The synthesis engine is near-certainly a wrapped third-party frontier model (it plugs into Claude/ChatGPT via MCP). The silence is the most telling omission in its own docs.
Barely. Experts are non-exclusive and can list on Delphi; MCP commoditizes delivery. The founders concede the failure mode: weak answers are “toothless AI slop,” no better than Googling. And Superhuman — whose CEO is a SuperMe angel — already pulled its own expert-clone feature after backlash and a class action.
Key Finding: SuperMe has top-tier backers (Forerunner, Greylock, Reid Hoffman) and genuine founder distribution chops. But the model it sells — real names on hallucination-prone output — pushes publicity, defamation, and consent exposure onto its users by contract while capping its own liability at $100, and the “intelligence” behind it isn’t SuperMe’s. Its founders are elite growth operators, not AI builders — strong go-to-market, thin technical moat.
SuperMe’s product is an authority-transfer machine: it lends a real expert’s credibility to AI output — then, in the fine print, hands the risk back to everyone but itself.
Public posts + connected Gmail, Docs, Granola notes train a named expert’s profile — sometimes from private correspondence.
An unnamed third-party frontier model generates answers “as” the expert. The intelligence isn’t SuperMe’s.
The answer is published under a real, identifiable professional’s name — including any hallucination.
Terms: output “may contain hallucinations”; users “MUST evaluate accuracy”; liability capped at $100.
Consent for every named person is pushed to the uploader; SuperMe keeps a perpetual, sublicensable license to the content.
The marketing and the legal documents describe two different products. The site says each profile “belongs to the person it represents” and “no profile exists without that person’s involvement.” The Terms take a “perpetual, irrevocable, transferable, sublicensable, royalty-free” license to that content and cap all harm at $100. Ownership is the pitch; a broad license back to SuperMe is the contract.
For a company whose product is AI, naming no foundation-model vendor — in marketing, Terms, or the Privacy Notice — is the single most telling omission. It means SuperMe’s cost of goods is someone else’s API and its core “intelligence” is rented. Because it also integrates into Claude and ChatGPT via MCP, the delivery layer is commoditized from both ends: it depends on the models above it and the protocol below it.
CEO; growth operator (Grubhub, Pinterest, Eventbrite), Reforge program partner — distribution, not ML
Forerunner, lead investor; frames the launch as “very timely”
Investor; reframes accuracy as a dial, not a bar — “20% better, 300% better or 10,000% better?”
Terms §16 limits liability to amounts paid or $100 — for harm from a named expert’s output
An angel’s own company pulled an expert-clone feature in 2026 after backlash + a class action
Coverage skewed skeptical; a Platformer parody compared it to “printing pictures of experts and letting you kiss them”
The expert-clone space is crowded, better-funded above it, and partly discredited behind it.
The closest direct rival — “digital minds” of experts. $16M Series A (Jun 2025, Sequoia lead; Anthropic’s Anthology Fund among backers). Experts can list here too; SuperMe’s roster is non-exclusive.
Consumer persona chatbots at a $1B valuation (a16z, 2023); Google paid a reported ~$2.7B for licensing + talent (2024). Dwarfs SuperMe on capital and model depth.
“Digital twin” personal AI; ~$11M across seed rounds. Adjacent to the same thesis, targeting individuals rather than an expert network.
The nearest precedent to SuperMe’s exact mechanic — an “Expert Review” AI-clone feature — was pulled after backlash and a class action. Its CEO is a SuperMe angel.
No lawsuit against SuperMe was found, and no substantive Reddit/HN thread surfaced. The exposure is forward-looking: one ugly misattribution involving a prominent expert — a false statement published under their real name — could trigger both legal action and mass expert churn.
Seven structural risks — several the company addresses only in the fine print, not publicly.
No owned model and no named vendor even in the legal docs. Margin (COGS is someone else’s API) and moat (the intelligence isn’t theirs) both sit on a third party — while MCP commoditizes the delivery layer.
Real names attached to hallucination-prone output, with liability contractually dumped on users ($100 cap). A hallucination voiced by a named expert’s profile is a false statement published under their name — and one ugly case could trigger mass expert churn.
SuperMe offloads consent duty to uploaders while taking a perpetual, sublicensable license. Live legal terrain (NY’s digital-replica law; the Lovo voice-clone ruling). The precedent feature was already pulled by an investor’s own company.
Experts are portable to Delphi; “a network of profiles” is replicable; MCP commoditizes distribution. The founders’ own “toothless AI slop” line names the failure mode.
Free today; the expert revenue split is admittedly unresolved; metering isn’t live. A two-sided marketplace needs both marquee experts and paying askers before the model turns on.
Elite growth and distribution founders, not AI/ML builders. That strengthens go-to-market but not the technical moat — which is exactly where the category’s risk lives.
Profiles train on Gmail “Sent,” Google Docs, and Granola notes. A profile built from one person’s private correspondence can surface third parties’ information — and consent for those third parties is pushed to the uploader by contract, not verified by SuperMe.
SuperMe is a strong distribution team selling borrowed authority on rented intelligence. The pitch is “ask the people who’ve actually done it”; the contract says the answer “may contain hallucinations,” caps the harm at $100, and never names the model doing the talking. The honest frame is not “AI experts on demand” — it is “real names on probabilistic output, with the liability routed to everyone but SuperMe.”
Based entirely on publicly available information, including the TechCrunch announcement of October 29, 2026. All confirmation is press- and company-page-level; no valuation, revenue, or user count is independently auditable, and SuperMe withholds its user count.
We build and maintain websites, help businesses of every size put AI to work, and run growth marketing that’s held to the same standard as our research.
