Creating a New Category With AI: What to Hand Over and What to Keep
What Lasts · Beyonds Journal · 2026-11-05
In 2004, 2 years before they founded HubSpot, now a large maker of marketing and sales software, Brian Halligan and Dharmesh Shah noticed a change in the way people shop and buy. "Buyers didn't want to be interrupted by ads, they wanted helpful information," the company's own history says. In October 2009 they published a book with Wiley, Inbound Marketing: Get Found Using Google, Social Media, and Blogs. Its description named the old ways, "cold-calling, email blasts, advertising, and direct mail," as "increasingly less effective," and its pitch fitted in one line: "Stop pushing your message out and start pulling your customers in."
HubSpot is a standard example of a company that created a category. Look at what stayed the same, though. A business still had to get customers to find it and choose it. What changed was the way, and HubSpot gave it a name.
For an AI founder today, this kind of move looks cheap: why not let a model find the gap and write the category? AI does some of that work well: it can sort what thousands of customers wrote and draft the text. The rest of the work decides whether a category holds, and it stays with you.
What does creating a category mean? April Dunford, who advises companies on positioning, defines it as the move "where you invent a new frame of reference for customers." In the language of this issue's Lead, a playground is every way one job gets done in a particular place, under that place's rules. Our reading: the founder names a new way of doing a familiar job, and the people on the playground decide whether the name holds.
Do category creators win?
The best-known case for creating a category comes from Play Bigger, a 2016 book by Al Ramadan, Dave Peterson, Christopher Lochhead and Kevin Maney. Its publisher describes "the new discipline of category design," by which companies "can create new demand where none existed." The figure behind it comes from the authors' own 2014 report: "Most Category Kings win 76 percent of the category's valuation." Measured across all the companies in their study, the 35 "category kings" held about 76% of the total value. The authors picked those kings by their own criteria, and they wrote that "The data on valuations and market cap are new and unstable."
76% — Share of total value held by 35 "category kings," leaders the authors chose themselves. Play Bigger Advisors, 2014
Dunford, in the same guide, sees a different pattern: "history teaches us that companies that create market categories often lose in the long run to companies that gained a market foothold after the hard work of creating the category was already done." Her examples are Google, which came after the early search engine Ask Jeeves, and Facebook, which came after Myspace.
Read together, the 2 views point one way: in Play Bigger's data most of the value goes to the category's leader, and in Dunford's experience the company that named the category often doesn't end up leading it. The founder who does the naming can end up building demand for someone else.
The founder who does the naming can end up building demand for someone else.
What did HubSpot, Drift and Gong actually create?
3 companies credited with creating categories each put a new name on a familiar job.
- HubSpot named its way "inbound marketing." The job: getting customers to find you and choose you. The old ways: cold calls, email blasts, ads, direct mail. HubSpot raised $5 million in its first venture round in 2007, 2 years before the book.
- Drift named its way "conversational marketing." The job: answering someone on your website while they are interested. A 2019 book by Drift's co-founder and CEO, David Cancel, and its vice president of marketing, Dave Gerhardt, described the old ways: "Online forms, email inquiries, and follow-up sales calls don't provide the immediacy that modern consumers expect." Drift had raised $47 million in its first 3 years, TechCrunch reported in 2018.
- Gong named its way "revenue intelligence." The job: knowing what actually happens in sales conversations, which teams had been recording incompletely. Norwest, one of Gong's investors, says Gong started with a $6 million seed round and that "Gong created a whole new category – revenue intelligence – and the market followed suit."
All 3 had investors' money early, and in each the job was old. HubSpot and Drift each explained the new name in a book, both published by Wiley. HubSpot's came 5 years after its founders noticed the change in buyers, and Drift's about 4 years after the company was founded in 2015.
Who decides what a category is called?
Gong's name shows how little of this the founder controls. In 2024, the analyst firm IDC published an assessment of "Worldwide Revenue Intelligence Platforms" and named Gong a leader. In December 2025, Gartner, another analyst firm, published its first Magic Quadrant (the firm's regular ranking of vendors) for a market it called "Revenue Action Orchestration". That market, Gong's release says, emerged from "sales engagement and revenue intelligence markets." In the same release, Gong's CEO calls the company "the creator and leader of the Revenue AI category."
Within about 2 years, the same work went by 3 names. Our reading: the founder proposes the name, and the people who buy and judge the product decide which one sticks. In Gong's case, analysts did much of the deciding.
The founder proposes the name, and the people who buy and judge the product decide which one sticks.
An AI-era name took another route. Clay sells software that combines data from more than 150 sources with AI research to personalize sales outreach. Around August 2023, as Clay began planning to sell through a sales team, Yash Tekriwal was taking 30 to 35 calls a week while cleaning data and building the systems behind them. On Gainsight's [Un]Churned podcast he recalled how his role got its name: Varun Anand, a co-founder, "just goes, I think we're going to call it GTM engineer."
When Clay raised 00 million in August 2025, it counted "over 280 GTM Engineer positions posted in companies like Cursor, Webflow, Notion, and Lovable." By June 2026, TechTree Research found 492 people with the title in Europe alone. Its report describes the work as "targeting, enrichment, outreach and routing," names Clay among the tools and doesn't say who coined the title. Before taking it, some of these people had held titles such as "Clay Expert" and "Clay Creator."
280+ — GTM Engineer positions posted when Clay raised its Series C. Clay, 5 Aug 2025
492 — People with the GTM Engineer title in Europe. TechTree Research, Jun 2026
Our reading: in Clay's case, employers did the deciding by hiring under the name. The job underneath, finding the right accounts and reaching them, is one sales teams have always done. A name that grows out of your own tool can carry the tool with it, as the earlier "Clay Expert" titles suggest.
What can AI do in this work?
Read what customers say, at a speed no founder can match. In a 2025 preprint, researchers from Cornell, Columbia and Waterloo used an AI model to sort Yelp reviews of Starbucks stores by topic (a task researchers call coding): 12,682 reviews drawn from a corpus of about 20,000. In a check against 300 reviews coded by people, they reported "high consistency between LLMs and human coders." "Human coders required a median of six minutes per review, whereas the LLM processed each in two seconds." It is one preprint, on coffee-shop reviews, and overall agreement fell short of the bar the authors themselves cite. It still suggests a founder can sort thousands of comments by complaint in an afternoon.
The places where customers write have rules of their own. G2, the software review site, updated its terms of use on July 9, 2026. Without G2's written consent, the terms forbid collecting its content "through automated, programmatic, or mechanical means" and using it "to train, test, validate, fine-tune, evaluate, or improve any machine-learning model, generative AI system." Reddit has required a contract for commercial use of its data since May 2024, and in October 2025 it sued Perplexity, an AI search company, and 3 data firms over scraping. The reviews themselves may be written by machines: in another 2025 preprint, people told real product reviews from AI-written fakes with 50.8% accuracy, close to a coin toss. The safest material is your own: interviews, support tickets and sales calls with people who spoke to you directly.
Source — Both studies in this section are preprints and have not yet been peer-reviewed.
AI can also draft a white paper. Google's guidance asks publishers "to manually factcheck and review all AI-generated content," and its spam policies target pages generated mainly to manipulate rankings, typically "large amounts of unoriginal content that provides little to no value to users, no matter how it's created." In our view, what makes a white paper worth reading is an observation the founder made in the field, and a model drafting from public text doesn't have one.
Isn't creating a category something only funded companies can afford?
Often, yes. "It's much more common for startups to start out positioning themselves in an existing category," Dunford writes, "until they have the money and momentum required to re-draw the lines around that category." Her alternative is a narrow version of an existing category: "we could win CRM for Investment Banks," CRM being software for tracking customers and sales.
Our reading: a new name earns its place when customers start using it to explain what they bought. A small company can get there in one place, with one kind of customer, long before it can afford a book or a conference.
What stays with the founder?
3 parts of the work stay with the founder.
The first is choosing the job and the place. A model can sort complaints. Deciding which job you will do differently, for which people and in which place, is a judgment about where you can win, and you are the one who carries its cost.
The second is seeing what people have learned to put up with. In 2004, HubSpot's founders saw buyers tiring of interruptions while marketing still ran on cold calls, email blasts and ads. Observations like that come from watching people do the job. A model reading old reviews sees only what someone thought to write down.
The third is writing the position in your own words. The sentence that says whose way of doing the job you change, and how, is a commitment. AI can tidy the sentence. Customers will hold you to it.
A test for this week: give your way of doing the job a name of no more than 3 words, and use it with 2 customers. Then ask each of them to describe your product as they would to a colleague. If they use your name, you have the start of a category. If they use their own words, keep theirs.
You can run this test alone, or with us and other founders in AI Startup Reality. From December 1, 50 founders spend 4 months building their AI companies with us, sharing their progress online, mistakes included. Applications close on November 27.
Route — AI Startup Reality: market pull first, then the company. From December 1, applications until November 27.
FAQ
What is category creation?
Category creation means giving customers a new frame of reference for a product, usually a new name for a new way of doing a familiar job. HubSpot's "inbound marketing" and Drift's "conversational marketing" are common examples.
Do companies that create a category win?
Play Bigger, a 2016 book on category design, rests on its authors' 2014 report: in their data, category leaders held about 76% of the total value, and the authors chose those leaders themselves. April Dunford, a positioning adviser, notes that companies that create categories "often lose in the long run" to those that gain a foothold after the hard work is done.
Can a startup create a category without funding?
The best-known cases had venture money early: HubSpot raised $5 million in 2007, and Drift $47 million in its first 3 years. April Dunford, a positioning adviser, writes that startups usually start in an existing category, or a narrow version of one, until they have the money and momentum to redraw it.
Can AI find a new market niche?
AI can sort large numbers of customer comments quickly: in one 2025 preprint, an AI model's coding of Starbucks reviews on Yelp largely agreed with human coders, though overall agreement fell short of the usual reliability bar. Deciding which job and place to build for remains the founder's judgment. G2, a software review site, forbids automated collection of its content and its use to train or improve AI models.
What part of category creation stays with the founder?
Choosing the job and the place to build for, seeing what customers have learned to put up with, and committing to a position in your own words. AI can sort customer comments and draft the text. Customers hold the founder to the position.
Sources
- HubSpot, Our story: link
- Brian Halligan and Dharmesh Shah, Inbound Marketing: Get Found Using Google, Social Media, and Blogs, Wiley, October 2009 (publisher description via Porchlight Books): link
- April Dunford, "A Quickstart Guide to Positioning," March 15, 2021: link
- Beyonds Journal, "How a Blue Ocean Is Created: Before the First Line of Code, Look at the Playground Your Product Is Aiming to Play On": link
- Al Ramadan, Dave Peterson, Christopher Lochhead and Kevin Maney, Play Bigger, Harper Business, June 2016 (publisher description via Porchlight Books): link
- Play Bigger Advisors, "Time to Market Cap" (market caps as of October 19, 2014): link
- HubSpot, "HubSpot Secures $5 Million in Venture Capital Funding Led by General Catalyst," September 12, 2007: link
- David Cancel and Dave Gerhardt, Conversational Marketing, Wiley, January 2019 (O'Reilly listing): link
- Katie Roof, "Drift raises $60 million to be an Amazon for businesses," TechCrunch, April 17, 2018: link
- Salesloft, "Salesloft Acquires Drift," Business Wire, February 13, 2024: link
- Norwest, Gong case study: link
- Gong, "IDC MarketScape Names Gong a Leader in Worldwide Revenue Intelligence Platforms 2024 Vendor Assessment," July 2024: link
- Gong, "Gong Named a Leader in 2025 Gartner Magic Quadrant for Revenue Action Orchestration," December 18, 2025: link
- Gainsight, [Un]Churned episode 197, "He Created a Job Title 1,000+ People Now Have ft. Yash Tekriwal (Clay)": link
- Clay, "AI GTM Leader Clay Raises 00M Series C to Fuel GTM Engineering Roles Industrywide," Business Wire, August 5, 2025: link
- TechTree Research, "The Rise of the GTM Engineer in Europe" (data as of June 24, 2026): link
- K. Boughanmi, K. Jedidi and N. Jedidi, arXiv:2510.16551 (preprint), October 2025: link
- G2, Terms of Use, last updated July 9, 2026: link
- Sarah Perez, "Reddit locks down its public data in new content policy, says use now requires a contract," TechCrunch, May 9, 2024: link
- Bloomberg, "Reddit Sues Perplexity, Others Over Alleged Data Scraping," October 22, 2025: link
- Meng et al., arXiv:2506.13313 (preprint), June 2025: link
- Google Search Central, "Google Search's guidance on using generative AI content on your website": link
- Google Search Central, Spam policies for Google web search: link
Beyonds Journal, issue 003
