Competitor Ad Intelligence

Overview

No public source publishes what a competitor spent on a creative. Every spend number in every competitor-intelligence product is a model output presented as a measurement.

So stop trying to measure spend and measure what is observable, and what a losing ad cannot fake for long: how long it has been running, how many variants exist, how many countries it runs in, whether the advertiser is accelerating, and whether it is still live.

score = 0.35·longevity + 0.25·variants + 0.20·geo + 0.10·velocity + 0.10·recency

When to use

• Building or reading a competitor creative corpus from any public ad library
• Deciding which competitor creatives are worth copying
• Explaining why a "top ad" list is misleading
• Finding the angles a competitor is winning with that you are not testing
• Setting up an ad archive that will still be useful in six months

Not for: estimating a competitor's budget. It cannot be done from public data, and a number invented for a client deck is worse than saying so.

Do lineage before anything else

An advertiser who re-uploads the same creative every fourteen days looks like three short tests. They are running one concept that has survived six weeks. Longevity is the heaviest signal in the score, so measuring it on ungrouped creatives produces a ranking that is confidently wrong.

python3 scripts/adscore.py lineage creatives.csv --out grouped.csv
python3 scripts/adscore.py score grouped.csv --top 20

On the sample corpus in templates/, the top creative changes identity once lineages are built. That is not a curiosity, it is the normal case.

Grouping rules: never across advertisers, token-set similarity on a normalised fingerprint at threshold 0.72, digits collapsed and punctuation removed. Raise the threshold on formulaic copy, lower it on long body text.

Stages beat scores when deciding what to copy

A battle-tested ad at score 60 is a safer bet than a new test at 75.

The output that changes what you do

python3 scripts/adscore.py gaps competitors.csv --mine mine.csv

ANGLE            THEIRS    SHARE     OURS    SHARE     GAP
how-to                1      33%        0       0%     +33%
offer                 2      67%        2      67%      +0%
urgency               0       0%        1      33%     -33%

Working for them, absent from your account: how-to

Not "here are their ads". "Here is the thing you are not testing." That is the only competitor output that reliably survives contact with a media plan.

Angle taxonomy, nine categories on purpose: offer, urgency, question, prize, curiosity, social proof, how-to, fear, novelty. Enough to be actionable in a brief, few enough that two people classify the same ad the same way.

Archive, do not query

The corpus is the asset, not the query.

• Mark disappeared ads as ended, never delete them. Deleting destroys the longevity series, which is 35% of the score.
• Store the creative file, not the URL. Ad image URLs expire; a library of dead links is a library of nothing.
• Store first seen and last seen per country, not just globally.

Nobody can sell you back a historical series you failed to collect.

Reference

Collection is out of scope, deliberately

This skill ships no scraping code. Use each platform's official API where one exists and its public interface where one does not, at a rate a human could plausibly produce, and read the terms of service first: they differ per platform, they change, and the consequences land on your account.

Common mistakes

• Scoring before grouping. The most common one, and it inverts the ranking.
• Quoting a spend estimate as a fact. It is a model output. Say so, or do not use it.
• Comparing scores across verticals. A 60 in insurance is not a 60 in fashion.
• Reading only text ads. Display and video carry the message inside the image; a text-only corpus is blind to most of what the competitor is saying.
• Copying the creative instead of the angle. The angle is what transfers. The creative belongs to their brand, their offer and their proof.
• Treating a keyword angle classification as final. It is a triage pass. Low confidence means unclassified, not "probably offer".