I Spent 25 Years Doing Competitive Intelligence by Hand. Here's the Honest Truth About Using AI to Do It.

For most of my career, competitive intelligence was something I did by hand.

Across 25 years in product marketing and go-to-market roles — at AWS, Microsoft, Symantec, and Citrix — I usually had the smaller-revenue products, the ones that fell outside the CI team’s coverage. Which meant I was the one who researched the competitors, wrote the battlecards, sat in on the competitive deals, and scrambled to keep all of it current every time the market shifted or sales asked me to cover another rival. I know exactly how much of a week disappears into that work, and how fast it goes stale.

So when general-purpose AI got good, I did what a lot of product marketers are doing right now: I pointed ChatGPT and Claude at the problem and hoped they’d give me my week back. Sometimes they did. Sometimes they handed me something confident, polished, and quietly wrong. This is my honest attempt to sort out which is which.

Where it genuinely helps

Let me start where the AI skeptics usually won’t: these tools are genuinely useful for competitive work — under one condition. You supply the raw material, and you already know the answer. If I understand a competitor and I just need the words — turn these notes into a talk track, tighten this objection response, draft the positioning paragraph — a general LLM is fast, flexible, and nearly free. That’s real work, and it saves real time.

It’s also the right tool when you’re only one or two competitors deep and a whole system would be overkill; when you want a thought partner to pressure-test your own thinking; or when it’s a genuinely one-off question you’ll never need to refresh. I still use AI this way every single week, and I’d tell anyone to do the same.

If that’s all you need, you don’t need to read the rest of this. Go save yourself an afternoon.

Where it quietly breaks

The trouble starts when you ask the AI to do the actual intelligence — not to polish what you already know, but to tell you what you don’t. That’s where I kept hitting the same walls, and not one of them was the kind you fix with a better prompt.

It retrieves; it doesn’t analyze. Ask for a battlecard and you get a tidy summary of the competitor’s website — the same things you’d find reading it yourself, stated more confidently. But knowing which of those facts actually matters in a live deal, which is a real threat versus marketing noise, and what a rep should do about it — that’s the job. And it’s the part a general model doesn’t do. What comes back reads like a feature comparison wearing a battlecard’s name.

It only researches the side you point it at. This one took me too long to notice. You can absolutely ask it to study your own company too — but the default ask is the competitor, nothing does your side automatically or keeps it current, and so your own strengths and weaknesses end up being whatever you happen to remember to type. A “why we win” built on one researched side and one half-remembered side isn’t analysis — it’s a guess with the competitor’s logo on it.

Your findings don’t compound. Yes, these tools remember things now, and my chat history is a scroll away — but that’s personal memory, not a team’s knowledge base. What I learned about a competitor last month lives in my account, not my colleague’s, and it’s a pile of conversations, not something organized, cited, and queryable that new findings automatically build on. So the work doesn’t accumulate into intelligence. It just piles up — and mostly I re-do it.

It doesn’t tell me when something changed. A competitor repriced on Tuesday, and I’d find out when a rep lost a deal on Thursday. AI research is a snapshot, and it looks exactly as confident on the day it goes stale as the day you made it.

Its sources are all-or-nothing. Lock it to sources you trust and it starves; open it to the open web and it will happily repeat a competitor’s own marketing as fact, or launder a single unverified post into a hard claim. There’s no dial in the middle — and in competitive work, that middle is where the truth usually lives.

And the output drifts. A battlecard I generated on Monday and a one-pager I generated on Thursday would quietly contradict each other, because nothing connected them. Then a rep is holding two documents that disagree, and trusts neither.

What I’ve learned to demand

After enough of this, I stopped asking “is the AI any good?” and started asking a better question: what does good actually require? Here’s the short list I’d hand anyone doing competitive intelligence with AI today — whatever tool you use.

• Sources you can click. Not “trust me.” A citation behind every claim, so checking it is one click, not a research project.
• Both sides, researched. If it hasn’t studied your own company with the same rigor as the competitor’s, its “why you win” is half-invented. Insist on two-sided.
• Something that accumulates. Findings should compound into a shared, living knowledge base your whole team can see — not evaporate when you close the tab.
• Change detection. Something has to watch your competitors and tell you what moved, or your intel is stale the moment you look away.
• Honesty about uncertainty. “Not stated” is not “no.” When two sources conflict, the disagreement should stay visible, not get quietly averaged into whichever version sounds best.
• A human in the loop. Nothing should reach a rep in a live deal that a person hasn’t approved. The trust comes from the sign-off — not from the model sounding sure.

None of that is exotic. It’s just the difference between a chat window and a system — and honestly, it’s most of what my job as a CI practitioner already was, long before I had software that could help.

The tool I wish I’d had

I believed in that list enough to go build it. After 25 years of doing this work by hand, I left product marketing to become a founder, and I built Signaro AI — an AI competitive intelligence program for smaller B2B GTM teams. It researches both your competitors and your own company, does the real analysis I used to do by hand (why you win, why you lose, how to handle the objection), cites every claim so you can check it, watches for what changes, and keeps a human in the loop before anything reaches your reps. It’s the tool I wish I’d had for two decades. We’re pre-launch and recruiting pilot users right now — so if you’re doing competitive intelligence with ChatGPT today and you recognized yourself anywhere in this, I’d love to show you the difference on a competitor you already know well. See it and join the pilot at signaro.ai.