Our own system · Competitor intelligence
Competitor Ads Intelligence
Competitor Ads Intelligence is an automated system that monitors every competitor's live Meta ads on a schedule, keeps a permanent record of every ad it has ever run, and shows the hooks, angles and offers your competitors keep coming back to.
Our own system, running today
40+ brands watched
The screen the team opens
One brand, in four numbers
How much we hold, how long it has been running, and what the system makes of it.
1 / 4
Competitor Ads Spy · one brand selected, read from the live records
Competitor Ads Spy · the captured ads table for one brand
Competitor Ads Spy · the intelligence report, as the system wrote it
Competitor Ads Spy · adding a brand to the watch list
What it runs on
The system reports back
We stopped checking. It tells us.
Every run writes to Slack. One message goes up when a run starts and the same message is rewritten when it lands, so a single line in the channel is the whole run, start to finish.
The second channel is the one we watch. It only posts when an ad appeared or an old one stopped. A quiet day there means nothing moved, which is a thing we used to have to go and check for ourselves.
The counts are the run. Fetched, inserted, updated, moved to inactive. If a brand has nothing live at all, the message says so rather than going silent and leaving us guessing.
Two channels, two jobs
The run log
One message per run. Posted the moment it starts, rewritten with the counts when it finishes.
1 / 2
Slack · the run log, one message per brand per run
Slack · the alert channel, only when an ad appeared or stopped
Research that went stale the moment it was done
Every advertiser's live ads are public. That sounds like an advantage until you try to use it. Someone opens the Ad Library, picks a brand, scrolls, screenshots a few ads and drops them in a folder.
Two weeks later that folder is a snapshot of a week nobody remembers. Which ads are still running? Which ones stopped? Which angle has been live for five months and which one lasted nine days? We could not say.
We wanted the opposite of a screenshot folder. Somewhere that keeps looking after we have stopped, and can tell us what moved while we were not.
One pass of the system
Competitor, scraped, stored, read
This is one run, start to finish. It takes about three minutes and nobody starts it.
A competitor goes on the list
One brand name and the link to its public Ad Library page. Nothing else to configure.
40+brands watched
What we point it at
A public wall, with machinery behind it
There is no private feed here. Our system reads the same page any person can open, and it reads it for every brand on the list, every run, without anybody deciding to look.
Behind that page sits the scraper. It is handed the same instructions every run, so a pull today and a pull in March collect the same thing.
That is the whole of what we point at a competitor. A public page and a fixed set of instructions.
Every run sends the same three instructions. Active ads only, every country, ad details on.
Meta Ad Library · one advertiser's public page
Why we built it this way
Four questions we kept re-answering by hand. Each one shaped a piece of what you have just seen, and none of them was answerable from a folder of screenshots.
What is this brand running right now? A scroll through a public page, on the day you happen to look.
What did they stop running? Invisible. An ad that disappears leaves nothing behind.
Which angle is actually working? The only honest signal in public is how long an ad has been live, and nobody was counting.
What is being said in the videos? Most of the spend is video. None of it was readable, let alone searchable.
How a run works
Six steps. A person does only the first, once per brand.
Add the competitor
A brand name and its public Ad Library link go into the watch list. Nothing else is configured.
Slack opens the run
"Competitor Ads Analysis started", the brand it is targeting, and a link straight to that brand's Ad Library page. The run is visible before it has a result.
The scraper reads the page
Apify returns every active ad the brand is showing in public, with the creative, the copy, the start date and the platforms.
Each ad is read once
The model turns one ad into an angle, an offer, a funnel stage, the triggers it leans on and an inferred persona.
The library reconciles
One ad, opened up
Open a row and you get the ad as it ran, then what the model made of it. Every field on this screen was written by a run. None of it was typed by a person.
A video ad and a static ad, same row, different evidence
A video ad, with the words that were said
The copy as published, the model's read on it, then the spoken script. We pull a transcript on demand, one ad at a time, from the Get button on the row.
1 / 2
Competitor Ads Spy · one captured video ad, expanded
Competitor Ads Spy · one captured static ad, expanded
The ad as it ran
Primary text, headline, call to action and landing page, as published.
What the model made of it
One angle, one offer summary, one insight line, and the triggers it leans on.
The video, in words
Pulled on demand, so a script can be searched instead of watched. 14 ads carry one.
Nothing invented
The read
Not one ad at a time. The whole set.
Reading one ad tells you what one ad says. The useful question is what a brand is doing across everything it has live, and which of those things it has committed to.
Asked
What is this brand betting on right now, and which of those bets has it proven?
This is the question our team asks before writing a campaign, and the one a client asks on a call. It used to mean opening the Ad Library, scrolling a brand's live ads and forming an impression from whatever was on screen that day. Now it is asked of any brand on the list, and asked of everything we hold on it.
BridgeA brand level read over every captured adThe model is handed every ad we hold for one brand, with its angle, offer, triggers, funnel stage and how many days it has been running. It is told to weigh repetition and longevity, and to treat a one off ad as a test rather than a bet. It writes one line, a short narrative and a list of bets back onto the brand's own record.
How we wired it
Public page in, one screen out
Nothing here needs access to anybody's ad account. Every layer reads what is already published.
What is already public
SourceEvery advertiser's live ads, on the page Meta publishes for them. One page per brand, and the link is the only thing we store about how to reach it.
Why any of this matters
An ad that has been live for 176 days is not a test.
It is the closest thing to a public admission that something works. 176 days is the longest run on file across every brand we watch, and the screen sorts on it.
What we learn
Hooks, angles, offers, and what a brand keeps repeating
This is the part we use. The system groups every captured ad by what it said, then counts how many ads carry each one.
A hook behind sixteen ads is a decision. An offer behind one is a trial. The count is the signal, and we do not do the counting.
Our team reads this before writing anything. Not to copy a competitor, but to know what has already been proven in the market we are about to speak into.
Counts move with every run. These are what the system held on this brand.
Four views of the same brand read
The hook bank
Every opening line the brand has used, and how many ads sit behind each one.
1 / 4
Competitor Ads Spy · hook bank, 35 hooks on one brand
Competitor Ads Spy · angle bank, 63 angles on one brand
Competitor Ads Spy · offer map, 63 offers on one brand
Competitor Ads Spy · the brand level read, as the system wrote it
The pieces, and what each one does
Five parts. Every one of them is something the team already pays for or already works in.
The public source
Where every ad on this page comes from
MetaAd Library, one page per advertiser
What runs, what stays human, what comes next
Here is the split. A system that watches other people's advertising is only worth trusting if it is straight about what it can see.
Runs itself
Running unattended
Five brands pulled every morning, one run a minute
Every new ad inserted, every returning ad updated
Ads that disappear recorded with an end date
The run log posting and then editing itself with the counts
A second channel woken only when something moved
You decide
Deliberately still human
What this system does not do
Our system reads public pages. It has no access to anybody's ad account, and nothing in it could change an ad even if it wanted to.
Everything it holds is something the advertiser chose to publish. We keep our own copy so we can tell what changed, which is the only thing the public page will never tell you.
No spend, no return on spend, no results. None of that is public, and the model is told not to invent it.
Audience and persona are inferred from the ad itself, not read from the advertiser's targeting.
The read carries an unverified badge on the screen, because it is a reading of public ads rather than a measurement.
Where the numbers on this page come from
Every figure here was read from the running system on 17 September 2026, not from a report about it.
The library counts
Read straight from the two tables the system writes to. 3,030 ads captured, 1,825 video, 2,202 with an end date, 828 still live. 2,360 carry a model written angle, 2,363 an inferred persona.
The run history
The 26 runs n8n still holds, across five days. 24 clean, 2 failed. A successful run takes 3m06s at the median, between 2m29s and 5m23s.
The Slack messages
Quoted from the workflow that posts them. A run log that posts on start and rewrites itself with the counts, and a second channel that fires only when a row was inserted or end dated.
What the team gets out of it
Nobody scrolls the Ad Library any more 40+ brands are on the list and they are pulled on a schedule without anybody deciding to look.
The history is the point 2,202 ads that are no longer running are still on file, with the date they stopped.
Silence means nothing changed A quiet alert channel means nothing moved, which used to be the one thing we could never be sure of. If a brand goes dark entirely, the run says so outright.
A hook with sixteen ads behind it is a fact The banks turn a pile of creative into counts, so we can see what a brand has stood behind more than once.
What we still do by hand
Short list, and we would rather it stayed short.
Choosing which brands go on the list
Asking for a transcript when an ad is worth reading in full
Reading the strategic summary and deciding whether we agree
Turning any of it into our own creative
Who this is for
The system is ours first. It is built to be handed over.
Brands that buy media and keep losing the thread on what competitors are doing
Agencies that do competitor research manually for every client, every month
Teams that already have a Slack channel where this belongs
Related resources
Know what your competitors are running.
We build the systems that watch the brands you care about, collect what they are advertising, and turn it into something your team can act on. Built on your stack, owned by you, running in weeks.
