Use AI for things only you can say. Stop using it for things anyone could say.
by Luis Gomes, Founder & Growth Lead
Here is the rule we use for every piece of content we ship — for ourselves and for clients. It separates the AI plays that compound from the ones that crash.
Use AI to move faster around things only you (or the client) can say. Don't use it to produce things anyone could say.
The rule is short, but the consequences of getting it wrong are big enough that they have already taken down sites. Some of those sites belonged to companies that thought they were following best practice.
What "things only you can say" actually means
Every brand has a narrow set of inputs nobody else has. Customer conversations. Internal performance data. The specific story behind the product. The methodology the operator built over years of running the work. The cohort behavior in the brand's actual database. The reason the founder started.
Wrap AI around those inputs and the output is unique by construction. AI is doing research, structuring, drafting, automating the parts that should be automated. The voice on the page is human. The substance on the page is proprietary. Nothing in the output is something a competitor could produce from the same prompt.
That is the play that compounds.
What "things anyone could say" actually means
Open ChatGPT. Type "write a 1,500-word blog post about [topic]." Read what comes back. Publish it.
That output is what the model has already seen across the web — recombined, smoothed, and presented in your brand's name. There is nothing in it that a competitor with the same prompt couldn't produce. There is nothing in it that the model itself doesn't already know. Worse: when the next version of the model gets trained, your post is in the training data, and the model will produce something even closer to your output the next time someone asks it the same question.
That is the play that crashes.
What's already happening
Lily Ray, who has been doing SEO for sixteen years and runs AI search for Amsive, has watched this pattern play out enough times to track it as a category.
She describes a "vicious cycle" in which low-quality AI content gets cited by other AI tools, which then becomes the training data for the next generation of AI content. Her own articles get paraphrased and her byline removed inside a week of publishing. The default ChatGPT model — used by something like 95% of users — does fewer fan-out queries and less background research than its predecessor, which means low-quality sources get cited and replicated faster than they used to.
The damage is not theoretical. Lily has personally watched 20 to 30 sites lose substantial traffic in the last three months because they leaned hard into scaled AI content. The pattern she calls "Mount AI" — a content spike followed by a crash — is the 2026 version of the Helpful Content Update pattern that wiped out small publishers in 2023. People refinanced houses. Closed shops. Some of those companies thought they were following best practices. They were following the version of best practices that worked the year before, in a category that turns over fast.
What works durably
The same Lily Ray interview names the plays that actually compound. None of them are surprising once you see the rule underneath them.
Original research with transparent methodology. Publish data that is yours. Show how you collected it. Be honest about the sample size and the limitations. AI cannot generate this — it can only summarize it after the fact, which is exactly what you want, because you are the source it summarizes.
Real expert humans visibly representing the brand. EEAT — experience, expertise, authority, trust — starts with experience first. If a real human with relevant credentials is publicly attached to the work, the brand earns the kind of authority signal that scaled AI content actively destroys. This is why the Leanboat blog has founder bylines and not "Leanboat Research" placeholder authors.
On-site copy that clearly explains what your brand is and what your team has done. Boring. Effective. The page that says clearly "this is who we are, this is what we have done, this is the experience we have" is the page that AI-search engines learn to cite as a primary source for your brand. The page that says nothing specific is the page that gets paraphrased by an AI tool and rendered as a generic answer with your brand omitted.
Engaging in real conversations, including in places that are hard to game. Reddit, LinkedIn comments, Facebook groups, niche communities. Not selling. Actually answering questions in the operator's voice. AI tools cite these places heavily because that's where firsthand experience still lives. Marketers are infiltrating them — Reddit is fighting back and mostly losing — but the principle holds: real human conversation is the part of the index that AI cannot synthesize away.
What we tell clients
Most clients want to be told to scale AI content. The cost-per-article math is irresistible. And the truth — which we say upfront — is that some of the tactics that scale right now will work for three to six months. Listicles where you put your own brand at #1. "Brand vs. alternative" comparison pages. Schema-stuffed feature comparison tables. They get citations in AI answers in the short term.
We are clear about the time horizon when we run them. They are not the moat. They are short-term tactics. The moat is the proprietary research, the expert voice, the on-site copy that clearly says what the brand actually is, and the real human conversations the brand is part of. The boring stuff. The stuff that compounds.
How we actually use AI
For the Leanboat blog and for client work, the workflow looks like this. AI handles the research — gathering sources, mapping the citation landscape, surfacing the angles the team has not seen yet. AI handles the structural work — outlines, fact-checking, source reconciliation, draft scaffolding. AI handles the production scale-up where the input is proprietary — generating multiple variations of a campaign brief from the same customer-data input, drafting personalized email tests against a real cohort, building dashboards from the data that is actually yours.
A human writes the final piece in their own voice. Always.
That is how content works in 2026. Use AI for the things only you can say. Stop using it for the things anyone could say.