AI for Competitive Intelligence Webinar March 26th

This session is about competitor analysis with AI: how to research your market and your rivals with deep research, how to pull competitive signal out of social media, and how to turn all of it into an automated dashboard that refreshes every single day. It is built for founders, consultants, and operators who want to know what their competition is doing without spending hours on manual monitoring. As always, the structure is quick wins first, then a live build, then demos of systems I actually run for myself and my clients.

How these sessions work (0:00)

Every week I take one use case or one precise theme and show how to implement generative AI inside of it. I do that in three ways. First, quick wins: prompts you can use directly in ChatGPT, Claude, Gemini, or any other provider. Second, a live build, where I take the use case and build something functional in about fifteen minutes. Third, demos of things I have built for myself or for customers, because a lot of people still use AI as a glorified Google when it is so much more than that.

Quick background on me: I studied astrophysics and have been living in France for nine years. When ChatGPT came out I started exploring business use cases for it, became the change manager for generative AI at Schneider Electric, and deployed AI upskilling for one hundred thousand employees as a team of three. Now I run my own business showing people how to use AI and helping them implement it.

Three channels for competitor analysis (2:53)

Today's topic covers three precise channels. One, deep research through the AI accounts you are already paying for. Two, social media: analyzing what your competitors publish and who engages with it. Three, websites: if a competitor communicates a lot through their site, you can track it with scraping.

For the quick wins, any of the big tools works, because we only need the deep research functionality that all of them ship. I use Claude for the demos, and for the building I use Claude Code. A lot of people have started using Cowork as well, and the gap between Cowork and Code is not really significant, so you can follow along with either. For the more advanced use case you will need an Apify account, the scraping platform I use to get data out of LinkedIn, YouTube, Instagram, and more.

Quick win: deep research for market analysis (5:17)

The first quick win is running a deep research analysis on your market. My prompt was simple: run a competitive analysis on the AI upskilling and AI implementation market for small and medium businesses in France, ideally the Parisian region, with the latest news on consulting firms in that market. You do not need to prompt it very specifically; the keywords are enough. If you want more accuracy, give it your website or a list of your services so it studies the trends directly related to what you sell.

Then select research mode. Deep research combines the reasoning capabilities of the model with web search and runs for several minutes, usually five to ten. Pick the biggest model available: for diligence and reasoning-heavy tasks, model size has a real impact on quality. Mid-demo I hit my usage limit and upgraded live to the two hundred dollar Claude Max plan, for the love of the game. The key point most people miss: you can schedule this as a task that runs daily, weekly, or monthly, and it becomes a recurring report on how your market is evolving.

The resulting market report named players I had never heard of, from big firms to boutique shops, with pricing for each, and flagged that the French government is injecting significant money into AI upskilling, which tells you where to position. Every claim comes with its source.

Researching a single competitor (9:30)

The second run targeted one company: my friend Theo's consulting firm, Omia (omia.io), because yes, we are competitors. The prompt: run a research analysis on this company, the people working there, any new hires you can find, product launches, and anything that could help me widen my services. Claude pulled from LinkedIn and Glassdoor, and the legal and company data came from Pappers, the French company-data site.

The report gave me the entire story of the company: when it was founded, who works there, what the website claims, how much they charge per day for AI training, consulting, and automation services, and even where they are vulnerable. It matched what I already knew from inside sources, meaning my friend. One practical tip: if you do not know your competition well, first ask AI for a report naming all your competitors, including ones you have never heard of, then run a deep research pass on each one. That builds you a positioning database.

Social media intelligence with Apify (13:09)

Social media is a gold mine, because if there is one thing people love to do, it is complain about things online. This works for any public profile on LinkedIn, YouTube, Instagram, or TikTok. For a company you get two sources: the profiles of the people working there and the profile of the company itself. Sometimes the company page is quiet, but the people still talk.

To extract the data I use Apify, a library of scrapers. In the demo I took the comments and engagements scraper for LinkedIn (15:45): paste the URL of a post and it returns the post content plus everyone who engaged with it. With that raw data you can detect whether your competitor signed a new client or hired someone new, because those things get announced in public posts. The idea is never to rely on one single source of truth. You want several sources, tracked over time, and this can run as a live system that updates every morning.

Live build: a competitor intelligence system with Claude Code (18:19)

For the live build I switched to Claude Code, already connected to the Apify platform, and built what I call a parasite system: something that feeds you information on any subject of your choice through a chosen channel. The build is a conversation. I gave it my friend's profile and asked it to fetch his last ten posts; it found the right scraper, quoted me a cost of about two cents, and returned the posts sorted by most recent. Then I asked it to do the same for his company page using the right Apify scraper (21:38), so both sources feed the same system, and you can combine that with the deep research reports from earlier.

From there the path is: build a dashboard on top of the data, deploy it to the web, and schedule it. If you are in a slow-moving industry, refreshing every seven days is fine; for fast ones, run it every twenty-four hours.

Keyword monitoring for market trends (27:08)

Sometimes you do not know exactly who to track, because your competitors and prospects are spread across platforms and languages. So instead of tracking a person, track keywords. My own example: there is a lot of buzz around Claude and around people migrating from ChatGPT to Claude. I built a system that searches LinkedIn every morning for posts containing the keyword Claude, then analyzes each one: is this a power user, someone actually migrating, or just noise?

The same data serves two purposes. It is a competition service, showing who does what I do, and it is a prospecting engine, surfacing people interested in those services. That is the pattern with almost everything you build with AI: the data source stays the same, and you change how you treat the data. Full disclosure from the live session: I hit the rate limit again despite the upgrade, so I switched to Gemini CLI mid-build to keep the demo going.

Deploying automated dashboards (31:29)

The dashboard I showed (53:59) is a plain HTML site deployed online. Every morning it loads the latest LinkedIn posts from the last twenty-four hours, qualified into power users, migrators, and noise, with the full content of each post attached. The stack is always the same: Apify in the back for data, Claude Code building the system, and one small service that turns it into a live platform. And funnily enough, mid-webinar someone messaged me asking for literally this same system on another theme; once you have it, multiplying it to new topics is copy and paste.

One technical note: Claude Code cannot do deep research natively, it does not have that functionality. The fix is to give it API keys so it can call the models itself and run the research inside your system.

Q&A highlights (36:00)

Can you trust the numbers deep research gives you? Every claim carries a source, and you can control the sources: list your preferred ones or blacklist the parasitic ones. If you want to go further, either verify manually or have a second AI go through the sources and verify them, which is a lesser-known but real use case. For source control, Perplexity is the best tool: under connectors and sources you choose whether knowledge comes from academic research, social media, or the web, and the Pro version refines that down to specific websites.

What about data privacy, do I need to build a RAG? In my opinion, no. There are three levels of privacy (41:30). Level one, the settings toggle: in Claude, go to Settings, Privacy, and turn off the option to help improve Claude. That is where most people stay. Level two, contact sales: in the European Union, providers must offer a zero data retention policy. Level three, local models with Ollama. Think of Claude Code as a robot: it has hands connected to your computer, but the brain, whether Opus, Sonnet, or Haiku, runs on someone else's servers. With a local model you throw away that brain, keep the body, and plug in a brain that runs on your machine. It is less powerful and needs a stronger computer, but nothing leaves your machine, and because you keep the body, the retrieval layer is already built, so there is no RAG to develop.

Claude Code versus Gemini CLI versus Antigravity? I used to recommend Gemini CLI as the budget version of Claude Code, but Google has been giving less and less usage despite advertising the opposite. Antigravity results have not been great lately either. Right now my recommendation is Claude Code as the main tool, full stop.

How do you keep Claude Code from wandering across connected tools? Organize your workspace. I keep my knowledge siloed with skills, each one pointing to one precise folder or data source. I do a workflow once end to end, then ask Claude Code to turn it into a skill so it is repeatable.

Wrap-up (56:39)

Today was mostly demos: deep research on markets and single competitors, engagement scraping with Apify, keyword monitoring, and a live-built intelligence dashboard. The next session takes the practical angle, where you build along with me. All the past and upcoming sessions are on the webinars page.