Claude watches how your competitors are using AI, and where a new AI-native player could move in on your space.
A free skill you build yourself in Claude Cowork. It wakes up once a week on its own: a short brief when something's changed, or a single line when all is quiet.
When you fill the profile, you set what should wake you and what shouldn't. The Watchdog stays silent until something clears that bar. Most weeks nothing does, and it says so in a single line rather than a brief you don't need.
You need one short profile: what your business does, the one differentiator you would hate a competitor to match, three to five competitors, your category, and your line on what counts as "material" (the AI moves big enough to be worth acting on). You have two ways to build it, and either way you save it in Claude as a file named "My Competitor Watchdog - profile".
This profile is what the whole thing runs on. The Watchdog reads it on every run, so the sharper your differentiator and your definition of "material", the sharper its briefs and the less noise you get. You only do it once.
Option A. Fill it in yourself. Copy the template below and complete each line. Ten minutes, and you know it is exactly right.
# My Competitor Watchdog - profile ## My business - What we do (one sentence): - Our single biggest differentiator (the thing we would hate a competitor to match): - Our market / category: - Where we mostly compete (region and segment): ## Competitors to watch (name 3 to 5) 1. 2. 3. ## Also watch the category for - New AI-native entrants or startups undercutting this category: yes / no - Specific AI capabilities I am worried a rival could add: ## What counts as "material" to me (so you only warn me for these) - A competitor shipping or announcing an AI feature that touches my differentiator - A competitor visibly adopting AI in sales, support, delivery or pricing - A new AI-first entrant in my category - A major platform or tool that lets my own customers do what I do, without me ## What is NOT worth warning me for - Generic press releases, funding with no product change, blog posts, awards, conference talks
Option B. Let Claude draft it for you. If you would rather not start from a blank template, paste the prompt below into Cowork. Claude researches your business and competitors and fills the profile in. Then you read it through, correct anything it got wrong, and save it. You stay in control, Claude does the first pass.
Research my business and draft my Competitor Watchdog profile. My business, in a sentence: [describe what you do] Using that, fill in a profile with these parts: - What we do - Our single biggest differentiator (the thing we would hate a competitor to match) - Three to five real, current competitors (find them, do not guess) - Our market or category, and the region and segment we compete in - What counts as "material" to me: any AI move that could genuinely threaten the differentiator above. That includes an established rival adopting AI, a new AI-native entrant undercutting the category, and a major platform or tool that lets my own customers do it themselves Then show me the finished profile so I can read it, correct anything you got wrong, and save it as a file named "My Competitor Watchdog - profile". Do not invent competitors or facts. If you are unsure about something, say so rather than guessing.
A project in Cowork is a folder Claude keeps for one job, with its own files it can read every time it runs. In Cowork, create a new project and call it Competitor Watchdog, then save the profile you filled in ("My Competitor Watchdog - profile") into it. That one file is all you need to add.
The Watchdog looks after its own memory. On its first run it creates a running log inside the project, notes what it told you each week, and reads it back before the next run, so it never repeats a signal you have already seen. That is what stops a monitoring tool becoming the noise it was meant to save you from. You never touch the log yourself.
A scheduled task is Claude running the same instructions on a timer, on its own, without you starting it. It is a Claude Cowork feature. Open the project you made in Step 2, then set the task up one of two ways: type /schedule in the project and follow the prompts, or open the Scheduled sidebar, choose New task, then Set up manually. Either way, point the task at your Competitor Watchdog project so it can read your profile, paste in the Watchdog prompt below, and set the frequency to weekly.
One thing to check: the task must be attached to your Watchdog project. If it is not, it still runs and still reports "all quiet", but it never sees your competitors, so you would not know it was empty.
Weekly is the right rhythm on purpose. Competitor moves land over weeks, not minutes, so a weekly check catches what matters without turning into a daily distraction. Once this is set, the Watchdog runs itself from now on. You never open it to trigger a run again.
You are my Competitor Watchdog. You run once a week, overnight, on your own. Your job is to watch HOW my competitors are using AI, and to warn me ONLY when something material has actually moved. Signal, not noise. I would always rather you tell me "all quiet" than pad a brief. STEP 1 — Read my context. - Read my saved file "My Competitor Watchdog - profile" for my business, my differentiator, my named competitors, my category, and my definition of "material". - Read my saved file "Watchdog - running log" if it exists. It lists what you have already reported, with dates. Note the date of your last run. Never report anything that is already in the log. - If either file is missing, say so plainly in your reply and stop. Do not guess my competitors. STEP 2 — Search the web, scoped to what is new since your last run. For EACH named competitor, and for the category as a whole, search for concrete signals of AI USAGE, not generic news. Look specifically for: - New AI-powered features, products or services they have shipped or announced - Public signs they are adopting AI in sales, support, delivery or pricing - AI-related job postings, as a proxy for where they are investing - A new AI-native entrant undercutting my category - A major platform or tool adding AI that lets my own customers do what I do, without me Ignore: press releases with no product change, funding with no product change, award wins, generic thought-leadership, conference talks, and anything already in my running log. STEP 3 — Judge materiality. Be strict. For each signal, ask: does this touch MY differentiator or MY category in a way that would matter to me, per my "what counts as material" list? Discard anything that does not clearly clear that bar. Most weeks, little or nothing clears it. That is normal and correct. STEP 4 — Decide the output. - If NOTHING material cleared the bar, reply with exactly one line: "All quiet this week. Nothing material moved on the AI front. Next check in 7 days." Then stop. Do not write a brief. - If ONE OR MORE material things cleared the bar, write the Watchdog Brief in STEP 5. STEP 5 — The Watchdog Brief (only when something moved). Keep it to a 40-second read. HEADLINE: the single most important thing, in one plain sentence. WHAT MOVED: 2 to 3 bullets. Each names the competitor, what they did, and a real source link. Facts only. If you cannot find a real source for a claim, leave the claim out. WHY IT MATTERS TO YOU: 1 to 2 lines tying it directly to MY differentiator or MY category. This is the part I care about most. THE ENTRANT ANGLE: one line on what a new AI-native competitor could now do to me, given this signal. Leave blank if not relevant this week. ONE MOVE TO CONSIDER: a single, specific action for this week. One move, not a strategy deck. Present it as a clean, well-formatted artifact: a bold headline, the sections clearly separated with space, and the source links inline. It should read as considered, not a wall of text. (This is my brief for my own business, so keep it plain and professional. Do not add any agency or vendor branding.) STEP 6 — Deliver it (optional). If I have connected a messaging tool such as Slack, post the brief to me as a formatted message: bold headline, the "what moved" points with their source links, why it matters, and the one move. Turn link unfurling off so the sources stay as tight inline links and do not expand into preview cards. Keep it scannable. If I have not connected a messaging tool, the brief simply waits for me in the app. Either way, on a quiet week send nothing beyond the single "all quiet" line. Never interrupt me for nothing. STEP 7 — Update the log. Append a dated one-line summary of everything you reported to my "Watchdog - running log" file, so you never repeat it next week. If the log does not exist, create it. SOURCING DISCIPLINE (read before writing any claim - this is how you stay accurate): 1. Open every source before you cite it. The exact claim must appear on the page you link. If you cannot open it, do not cite it. 2. Prefer the company's own announcement or official blog over any secondary blog or aggregator. Cite the primary source. Use a reputable outlet only when no primary exists, and say so. 3. One announcement is one signal. If many outlets cover the same thing, cite the single most authoritative one, do not count it as several moves. 4. State a number only if that exact figure appears in your source. Never round, estimate, or infer a statistic. If the source says "about a third", do not write a precise percentage. 5. Name every product exactly as its source names it. Never approximate or invent a product name. If you are not sure a name is real, drop the claim. 6. Say "announced" for things not yet shipped, and "reported" for anything you could not confirm from a primary source. Do not state either as settled fact. 7. Date every signal, and only report what is new since your last run. 8. Before you send, re-read every claim and check three things: did I open the source, does the exact claim and number appear there, is the product name real. Drop or soften anything that fails. At the foot of the brief, note how many sources you read and how many you cited. 9. If a claim cannot pass these checks, leave it out. A wrong brief costs more trust than a quiet week. RULES: - Never fabricate a signal, a feature, a source or a number. If you are unsure, leave it out. A false alarm destroys my trust in you faster than a missed signal. - Prefer silence to noise. "All quiet" is a valid and often correct answer. - British English. Plain language. No jargon, no hype. Write like a sharp colleague, not a newsletter.
Set the task to skip approvals so it can run while you are away. Approvals are the checkpoints where Claude pauses to ask your permission before each action. Skipping them lets it run start to finish on its own. Claude will show you everything it could reach and warn you. It is safe here for one reason. This task only searches the web and writes to its own files in the project, your profile and its log. It is never told to open your email or anything else, so it does not. Pause and think before you skip approvals on a task that touches your inbox.
With this switched on, the Watchdog can finish a full run on its own while you are away, and the result is waiting when you next open Claude.
That is it. The task is now live and will run every week on its own. Next time you open Claude, either a short brief is waiting or a single line tells you all is quiet. Both are the system working exactly as it should.
Your brief waits for you inside the task, ready to read the next time you open Cowork. The prompt above also asks Claude to post it to Slack if you have Slack connected, so it can come to you instead of waiting.
From now on, something is quietly keeping watch on what your competitors are doing with AI, and it only speaks up when it is worth your attention. You do not think about it again until it has something to tell you.
A brief is a 40-second read. One headline. What moved, with a real source link on every claim. Why it matters to you. What a new AI-native rival could now do. One move to consider this week. Here is a real one, from a run against a marketing agency profile.
You just built Competitor Watchdog. Here are two more free, build-it-yourself resources from the Serpin team.
Sales Recon is Julia's skill that sends six AI agents to research any prospect and writes you a sourced one-page brief before your call.
First Playbook is Scott's guide for leaders choosing where to start with AI agents. The first one you build sets the trajectory, so it walks you through picking one that ships and pays off.
