There is a particular kind of professional honesty that emerges when someone admits they have tested every major AI tool on the market and concluded that none of them deserve her money. Livia Hirsch, a B2B content strategist who helps scale-ups crack the US market, has arrived at precisely this position. She uses Claude, ChatGPT, and Gemini. She pays for none of them. The free tiers, she has decided, are entirely adequate for the only thing AI actually does well in her workflow: being a thesaurus with delusions of grandeur.
Where AI Earns Its Keep: Word Choice and Idea Stress-Testing
The use case Hirsch describes for AI is so modest it borders on insulting to every venture capitalist who has poured billions into the technology. She gets stuck on a word. She has written “however” three times in three paragraphs and the alternatives feel wrong. So she asks an AI to generate fifteen or twenty options, picks the one that works, and moves on with her life. That is the entire value proposition. A brainstorming partner, she calls it, which is a generous framing for what amounts to a very expensive synonym finder.
She has also used AI to draft reusable brief templates for clients and to rewrite the occasional email in a more professional tone. These are tasks where the output is raw material, not a finished product. Everything gets edited. Everything gets rewritten. The machine suggests; the human decides. Nobody, Hirsch seems to understand, would confuse a thesaurus for a ghostwriter – though apparently several billion dollars of investment capital has been wagered on exactly that confusion.
There is a secondary use that proves slightly more interesting. She will sometimes feed a strategic idea into Claude and ask it to “poke holes” in her thinking, to stress-test a concept before presenting it to a client. This at least approaches something resembling intellectual utility. But even here, the tool is a sounding board, not a strategist. The ideas are hers. The critique is a prompt away from sycophancy.
Where AI Falls Apart: Writing Actual Content
Hirsch has tried. Obviously she has tried. She has fed AI brand guidelines, tone of voice documentation, specificity requirements – the full briefing package that would equip any competent freelancer to produce usable work. The machines produced something grammatically immaculate and substantively hollow. She describes reading AI-generated blog posts and encountering sentences that are structurally correct yet say absolutely nothing. “You’ve strung together a grammatically correct sentence that adds nothing,” she observes, which is perhaps the most precise autopsy of AI content anyone has yet performed.
The problem is not that AI writes badly in any obvious way. It does not misspell words or mangle syntax. The problem is that it writes the way a student pads an essay to hit a word count – every sentence technically present, none of them earning their place. Hirsch notices this because she pays attention to words, which is, she concedes, an occupational hazard. But it is also the reason her clients hire humans in the first place. Brand voice is not a set of rules you can feed into a prompt. It is a sensibility, a pattern of choices that AI can mimic on the surface without ever understanding underneath.
And the subscription question answers itself. If a tool cannot write and cannot research – which, as we are about to discover, it emphatically cannot – then the free tier is more than sufficient for brainstorming synonyms. Hirsch has Gemini bundled with Google Workspace. She barely uses it. The economics of paying for what amounts to an intermittently useful word generator simply do not add up.
The 45-Minute Research Disaster
The single most damning story Hirsch tells involves a McKinsey citation. She needed a specific statistic linked to its original source – a McKinsey report. This is, on paper, the exact kind of task that a system trained on the entire internet should handle with contemptuous ease. It did not.
The AI returned a 404 page. She flagged it. The AI apologized and offered a blog post that repeated the statistic without sourcing it. She flagged that too. The AI, apparently running out of ideas, linked her to Google – essentially telling her to do her own research. This loop continued for forty-five minutes. Forty-five minutes of progressively more creative failure, producing exactly zero usable citations.
“I’ve actually done this for 45 minutes and gotten no source. I should’ve just Googled it myself.”
The kicker is that she has repeated this experiment, by her own admission, “a stupid amount of times.” Across Claude, ChatGPT, and Gemini, the result is consistent. AI cannot reliably trace a statistic to its primary source document. You cannot cite McKinsey without a McKinsey link, as Hirsch puts it, which seems like a fairly low bar for a technology that is supposed to be revolutionizing knowledge work. Good old-fashioned desk research, she concludes, still wins. This is less a ringing endorsement of human capability than a devastating indictment of machine limitations.
Claude vs. ChatGPT: The People Pleaser Problem
Between models, Hirsch has developed a clear hierarchy based on a criterion that would make most AI product managers uncomfortable: honesty. When she asks Claude to critique an idea, it pushes back. It identifies weaknesses. It behaves, in other words, like a useful interlocutor. ChatGPT, by contrast, defaults to enthusiastic agreement regardless of what you tell it.
Hirsch describes asking ChatGPT to push back and receiving something that manages to be simultaneously critical and fawning – the conversational equivalent of a waiter who tells you everything on the menu is their favorite. “ChatGPT is such a people pleaser,” she says, and “it takes it to extremes.” Claude, she notes, is “a bit more steady,” which in context reads as the highest possible compliment: it will occasionally tell you that your idea has problems.
This distinction matters more than it might appear. A brainstorming partner that agrees with everything is not a brainstorming partner. It is a mirror with a smiley face painted on it. Hirsch needs friction – something that identifies the weak points in a strategy before a client does. Gemini, despite being Google’s product and theoretically better connected to search results, has not proven more capable at research. She uses it rarely, which is Google’s problem to contemplate.
What Her Clients Are Doing With AI
Among Hirsch’s enterprise clients, two contrasting strategies have emerged. One has built a custom GPT locked within their own ecosystem, using AI as an internal tool on their own terms. The other has made the deliberately contrarian bet of leaning away from AI entirely – investing in interviews, original quotes, infographics, and unique angles specifically to differentiate from the rising tide of machine-generated sameness.
Neither client uses AI for content writing. This is worth pausing on. These are not Luddites or technophobes. They are serious B2B operations that have evaluated the technology and concluded it belongs in coding, SEO, or process automation – not anywhere near the content that represents their brand to the world. One is building walls around AI. The other is building moats without it. Both have decided that the one thing AI definitely should not be doing is writing the words their customers read.
For the full interview breakdown, see our complete Expert Insight with Livia Hirsch.
Tools Mentioned in the Interview
The following tools and platforms were referenced during this conversation.
Related Content
B2B Content Creation Behind the Scenes: Expert Insights with Livia Hirsch
Livia Hirsch, a B2B content strategist who helps scale-ups expand from local to global brands, breaks down the tools, workflows, and hard-won lessons that separate content that converts from content that just exists.
Written by Sophie SteffenContent Metrics That Actually Matter in B2B
There is no single metric that matters in B2B content. Livia Hirsch explains which dashboards she builds, which numbers she ignores, and why 3 LinkedIn likes are not a crisis.
Written by Sophie SteffenThe B2B Content Tech Stack That Actually Gets Used
Livia Hirsch runs her entire B2B content operation on Miro, SEMrush, Google Docs, and Notion. The rest is whatever the client already has.
Written by Sophie Steffen