AI Thought Leadership Without the Noise

AI thought leadership has a supply problem. As I see it, there are about five topics in open circulation: agents, open versus closed models, safety, the cost of inference, and whether this is a bubble. A large share of the industry's executives seem to be publishing on them at the same time.

In my experience, most of that output is a summary of consensus with a company logo on it. It gets published, draws little response, and confirms a quiet suspicion inside the company that thought leadership doesn't work.

It works. What doesn't work is commentary, and the two get confused constantly. AI thought leadership is a published position on an unsettled question in AI, one an informed reader could argue with, from someone whose work gives them standing to hold it.

Key takeaways

• The test for thought leadership is whether an informed reader could argue with a specific sentence in it. A piece nobody could object to is a summary of consensus, not a position.
• Sauce Labs, a software testing company, announced its 2025 report with a press release led by a hard number: in a Wakefield Research survey of 400 US tech professionals, 95% reported setbacks from their organizations' AI initiatives.
• A column through a paid membership program such as Forbes Technology Council carries an annual membership fee under Forbes Councils' own terms. It's publishing through a paid membership, not earned coverage.

What's the difference between thought leadership and commentary?

Commentary explains what happened. Thought leadership asserts something that isn't yet settled, in a way that lets people disagree with you specifically.

The test is whether an informed reader could argue with it. If your piece on AI agents could have been written by any of forty companies and nobody could object to a sentence in it, you've written a summary. Summaries are free to produce and worth what they cost.

The uncomfortable part is that a real position exposes you. That's the mechanism. A claim that carries no risk of being wrong carries little information, and readers often sense that even when they can't articulate why the piece felt empty.

Three things a real position needs

A credible position in AI rests on standing, on specificity that costs you something, and on data nobody else has.

Standing. You have to be entitled to the claim by what you've built or seen. An inference company can speak about production economics from its own numbers, which a commentator working secondhand cannot. A vertical software company can speak about what AI does to a specific workflow from what it has measured. Standing is the constraint most companies ignore, and it's why executives get pushed to opine on general AI policy when their actual expertise is narrower and more interesting.

Specificity that costs you something. Naming what your approach is bad at, or what you got wrong, buys more credibility than any positive claim. Reporters have been told for years that AI systems work flawlessly. The omission has become the tell, and its inverse has become a fast signal that you're worth listening to.

Data nobody else has. The strongest position draws on data only your systems produce, published with enough about your method and its limits that others can assess the claim even if they can't reproduce the data. Research you commission can do the same job when you publish how it was done. In my view production data is the most underused asset in AI. Plenty of companies can see in their own production data what models actually cost, where they fail and how usage shifts, and they publish opinion pieces anyway.

Which AI thought leadership topics are still open?

Not every corner of AI thought leadership is saturated equally.

The crowded ground, in my view, is anything about model capability, agent futures, or whether the industry is in a bubble. Countless voices, no barrier to entry, and the loudest participants already have larger platforms than you will build.

The more open ground, as I see it, is your own operating experience: how your team decides a model is good enough to ship, what broke after launch, what a workload cost once real users arrived. Surveys already cover these questions in aggregate. A company describing its own answers is rarer, and in my experience the questions are unglamorous enough that few try. An operator answering them from experience has something a commentator can't fake. Being cited for that kind of answer compounds in a way that being published does not.

The other open ground is vertical depth. What AI actually does to construction scheduling, or software testing, or payments risk, gets less firsthand explanation than it deserves, in my experience, because the people who know are busy running it. A vertical company that can explain a specific workflow, what it measured and where the approach failed, competes for attention with far fewer voices than a company commenting on AI in general.

Sauce Labs, a software testing company, is one example from our own work. The 2025 report we launched with it was built on a Wakefield Research survey of 400 US tech professionals. The press release announcing it led with the finding that 95% of respondents reported setbacks from their organizations' AI initiatives. That is a vertical company putting numbers behind an arguable claim about what AI is doing to its own field.

When Cursor's AI support agent started making up a policy that didn't exist, Sauce Labs caught it as the story was breaking. We pitched everyone covering the space right away, and within a few hours their spokesperson was briefing the New York Times and Bloomberg, and The Register and Ars Technica followed before the cycle moved on. It worked because they had a clear point of view on it, so they were part of the story, not commenting on it after.

Formats, ranked by what I've seen them return

Original research and data. In my ranking, the highest return and the highest effort. Good research gives other writers evidence to cite, and a citation keeps working long after an impression fades. When we design survey research, we write the questions so either answer is a story, asking what's stopping adoption rather than whether companies are adopting. We publish the sample, the audience and who paid, because in my experience reporters check all three.

Contributed articles and bylines in trade press. Reliable, and undervalued because the outlets are less glamorous than the general tech press. A byline in the publication your buyer actually reads outperforms a mention in one they don't. One distinction to keep straight: a column through a paid membership program, such as Forbes Technology Council, comes with an annual fee. It can be useful distribution, and Forbes Councils editors review submissions, but it's publishing through a paid membership rather than earned coverage, and it should be identified that way.

Podcasts. Good for depth and for building a spokesperson's reputation over time. Immediate reach is usually modest unless the show itself has a large audience. Worth doing when you're building a person, not a moment.

Op-eds in business press. Hard to place, high impact when placed, and only viable when you have a position on something genuinely contested and the standing to hold it.

LinkedIn posts. Fine as distribution, and useful for discussing ideas with buyers directly. In my experience they aren't a strategy on their own, and treating them as one is how a program stalls out with lots of activity and no outside validation.

The failure mode to avoid

The accountability reporters on the AI beat cover harms, labor effects, environmental cost and the gap between claims and reality. They are not a distribution channel for a company's perspective, and pitching them thought leadership as though they were is the fastest way to attract scrutiny you didn't want.

I've written about when they are the right call in Who Actually Covers AI Companies. For thought leadership, the test is internal: before you publish a position that invites their attention, be honest about whether your own practices could withstand the examination.

How should an AI company start on thought leadership?

An AI company can start thought leadership with one recorded conversation rather than a blank page. When we build a program, it begins with a 45-minute interview with the executive, and I open by asking them to skip anything bland. I'd rather hear a strong, counterintuitive opinion or a concrete customer example than vague generalities.

Pick the narrowest question you have genuine standing on. Write the version that could be wrong. Attach data only you have. Then place it where your buyer reads rather than where your competitors will see it.

That last distinction is the one I see programs get backwards most often.

Common questions

What is AI thought leadership? Publishing a defensible position on an unsettled question in AI, backed by standing and ideally by proprietary data. It's distinct from commentary, which explains what already happened and rarely produces a response.

Does thought leadership actually generate business? It can, mostly indirectly, by helping prospects judge a company's expertise. In my experience the return shows up as inbound from people who read the piece, as credibility in sales conversations, and as citations that keep circulating, and it takes longer to show than an announcement does. It's a poor substitute for news coverage and a good complement to it.

Has AI-generated content made thought leadership harder? In my view it has made summaries worth less and positions worth more. Anyone can now have a model draft a competent overview of agents or inference costs, so a piece that only explains is easier than ever to ignore. What a model can't produce for you is standing: what your team measured, decided and got wrong. That's where the effort should go.

What topics should AI companies avoid? Broad commentary on model capability, agent predictions and bubbles, unless you have firsthand standing to add something. In my view that's where competition is heaviest. A specific question your team can answer from its own work, about running these systems in production or about an industry workflow, is usually a better place to start.

Who should be the thought leader? Whoever has genuine standing on the specific question, which frequently isn't the CEO. Technical founders and operators tend to have more credible positions on narrower topics, and narrower topics are where the openings are.

Sources

• Forbes Councils, Our Terms of Service (Membership Terms of Use and Agreement), last revised June 11, 2024
• Sauce Labs, Sauce Labs Report: AI 'Gold Rush' Has Decoupled Speed from Quality, Leading to Gaps in Both Talent and Tooling, ACCESS Newswire, September 17, 2025

Salient PR works with AI infrastructure and AI-native companies including Tenstorrent and Browserbase, building spokesperson programs and contributed content across technical and business press.