The Prompting Company: AI discovery is not an industrial quote

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A manufacturer can have a strong product and still be absent when a buyer asks ChatGPT, Claude, Gemini or Perplexity what to shortlist. The Prompting Company addresses that discovery gap. Its public offer is not an autonomous industrial buyer: it is a GEO (generative-engine optimisation) layer for question research, AI-readable content and agent-facing page routing.

What the company actually sells

The Prompting Company, Inc. is a San Francisco startup founded by Kevin Chandra, Michelle Marcelline and Albert Purnama, listed by Y Combinator as an active Summer 2025 company. The founders previously built Typedream and Cotter, which were acquired by beehiiv and Stytch respectively. 1

The public workflow identifies high-intent questions, measures whether a product is mentioned, creates AI-optimised content and routes crawlers to a clean Markdown version of a page. The Starter plan is publicly priced at $300 per month and includes question discovery, citation, frequency and sentiment analysis, and agent routing. Pro is $1,500 per month and adds eight AI-optimised articles per month. Setup, usage, integrations, exports and enterprise terms are not disclosed. 2

The company’s own explanation distinguishes discovery from usability: an agent must be able to find documentation and perform a workflow, not merely mention a brand. Its positioning has broadened from ChatGPT mentions to AI search, coding agents such as Codex and Claude Code, and personal agents. That broader framing still does not establish live ordering, quote approval or access to a purchasing API. 4

What the named cases report

In its NVIDIA case, The Prompting Company benchmarked answers about infrastructure cost, performance, return on investment and deployment. It added direct answers, definitions, comparison tables, frequently asked questions and cost-per-token evidence, with NVIDIA experts and partners validating assumptions. The provider reports cost-efficiency framing rising from 33% to 48%, NVIDIA Blackwell mentions from 26% to 45%, and CAPEX/OPEX framing falling from 23% to 17%. 5

Those are provider-reported answer-framing changes, not audited sales or return-on-investment results. The case supplies no prompt count, model list, dates, denominator, sampling method, confidence interval or independent corroboration.

The Hyperbrowser case describes questions about browser use by AI agents, concurrency, stealth scraping and Playwright migration. It reports share of voice increasing from 3% to 58%, with Hyperbrowser becoming the number-two recommended platform. Again, the prompt count, model versions, period and denominator are undisclosed. The result is useful as a vendor case claim and possible test design, not proof that visibility caused revenue, adoption or procurement wins. 6

TechCrunch reported in October 2025 that named customers included Rippling, Rho, Motion, Vapi, Fondo, Kernel and Traceloop; that a Fortune 10 company used the product; that about 500,000 pages were hosted; and that platform-driven client traffic reached the double-digit millions per month. These were company-reported figures in a founder interview. The same report said agents were then focused on discovery and recommendation rather than clicking options or accessing application programming interfaces directly. 7

An independent October 2026 review reported that its own name queries in ChatGPT, Claude and Perplexity did not identify the company, while Gemini found an unrelated UK creative agency. That is a dated challenge, not a definitive failure: the review does not publish a reproducible prompt corpus, model versions or controls. 8

An illustrative industrial application

For a manufacturer, dealer or service workshop, this could be an illustrative discovery and content pilot, not a reported industrial deployment. Start with controlled capability pages covering equipment limits, materials, tolerances, applications, drawings, revision dates and project evidence. Separate an indicative estimate from an approved binding offer. Keep the ERP, customer relationship management, engineering review and RFQ or configure-price-quote process authoritative for feasibility, capacity, inventory, compliance, price and acceptance.

Pre-register buyer questions across the relevant models. Log the answer, citations, incorrect claims, abstentions and correction owner. Measure qualified inquiries and factual accuracy against a dated baseline, not just mention share. Generated content must not authorise an order, substitute a material, duplicate outreach or expose restricted customer and project data.

Funding and limits of the evidence

Public reporting describes one $6.5 million seed round in October 2025, led by Peak XV with Base10, Y Combinator, Firedrop and angels including Logan Kilpatrick. The retained funding records contain repeated CSV entries and differing announcement dates; they should not be summed as multiple rounds. 7 9 10

No retained evidence establishes a conversion, qualified-lead, adoption, revenue, return-on-investment or total-cost result. There is no public API, integration/export specification, permission or data-retention model, Europe-specific availability, language coverage or industrial workshop case. The practical lesson is therefore bounded: improve the evidence available for AI discovery, but keep discovery separate from a sale and preparation separate from commercial authority.

Sources and next reading

• Official product and pricing — public overview and prices.
• Customer stories — public story index and agent-experience FAQ.
• NVIDIA case — provider-authored case study.
• Hyperbrowser case — provider-authored case study.
• Y Combinator company profile — public company profile.

Explore OfferSignal · Focused Business & AI Workshop · All research

References