# Advertising intelligence: what it is and what the data can prove
> Advertising intelligence turns public ad data into budget, creative and positioning decisions. Here is what each source exposes and which metrics hold up.
- **URL:** https://adextract.co/blog/advertising-intelligence
- **Published:** 2026-11-20
- **Author:** Gaurav Singh
---
Advertising intelligence is the practice of collecting public ad data from platform libraries, turning it into decisions about budget, creative and positioning, and discarding anything the data cannot support. It answers four questions: who is advertising, what they are saying, where the money is moving, and what changed since last week.

## What is advertising intelligence?

The discipline has three parts, and most teams only build the first. Collection pulls ads, placements and estimated spend from public libraries and licensed APIs. Interpretation turns those observations into a claim about intent, such as a competitor moving budget from prospecting to retention. A decision protocol says in advance what would make you act, so the claim lands on someone with authority to change a campaign.

It is not an ad spy subscription, and it is not a creative swipe file. Both of those are collection tools. A swipe file tells you what a rival ran. Intelligence tells you what changed, why it probably changed, and what you will do about it before the next flight of creative goes live. That distinction is the whole job, and it is what separates a dashboard people screenshot from a report that changes a budget line.

## The four layers of an advertising intelligence system

Build the layers in order and the system holds. Build them out of order and you end up with a folder of screenshots nobody opens.

- Layer one, collection: scheduled pulls from ad libraries, transparency archives, landing pages and licensed APIs, stored as snapshots rather than screenshots.
- Layer two, normalisation: one record per creative, deduplicated across ad flights, with the same field names whatever platform it came from.
- Layer three, interpretation: a written claim about what a change means, tied to a specific competitor and a specific date.
- Layer four, decision protocol: pre-agreed thresholds, an owner, and a review slot on the calendar where the claim gets accepted or killed.

Layers one and two are engineering. Layers three and four are management, and they are where nearly every underperforming programme fails. A team with a spreadsheet and a Thursday meeting beats a team with a licence and no meeting.

## What does each public data source actually expose?

Coverage is not uniform, and pretending otherwise is how teams build a confidence they have not earned. Here is what each of the free surfaces really returns in 2026.

The Meta Ad Library shows active ads across Meta apps, plus every ad that has run on a page for commercial advertisers in most markets. Meta states that ads about social issues, elections or politics carry spend, reach and funding entities and stay in the library for seven years, while ads that delivered an impression in the EU are archived for one year after their last impression (Meta Transparency Center, ad library tools page, updated August 24 2023). Commercial creative outside those categories is observable as creative, not as spend.

- Meta Ad Library: creative, start dates, page identity, and whether an ad is still active. No spend, no targeting, no delivery volume for ordinary commercial ads.
- Meta Ad Library API: political and social-issue ads, EU-delivered ads, and branded content. If your competitor is a US ecommerce brand, the API does not cover their commercial creative. Plan for that rather than discovering it in week three.
- Google Ads Transparency Center: advertiser-level lookup with creative, format and first-seen dates. No keyword or copy search, and no spend figure.
- TikTok Creative Center and the Commercial Content Library: trending creative, top ads by market and category, and EU-facing ad transparency. Useful for formats and hooks, weak for tracking a named competitor's budget.
- Licensed ad intelligence APIs: modelled spend, impression ranges, placement logs and creative archives across networks in one schema. You are buying coverage and history, not ground truth.

The practical rule: judge a source by the question it can answer. Creative longevity needs a library. Spend direction needs a model. Keyword intent needs auction and transparency data. One surface that claims all three is either expensive or guessing. Our breakdown of the ad intelligence API against the Meta Ad Library API digs into which of the two you should actually wire into an agent.

## Why every competitor spend number is a model, not a receipt

Nobody outside a media agency's finance system sees another advertiser's invoice. Every competitor spend figure you will ever read is inferred, usually from observed impressions multiplied by a rate card range, adjusted by season and format. The estimate can be useful. It cannot be a receipt, and no amount of decimal places changes that.

The political category is the clearest evidence, because there the numbers are audited and still move. AdImpact reported $12.06 billion of political ad spending across the 2025-26 cycle, the first midterm cycle to outspend a presidential one, and more than $460 million above its own June projection. Broadcast took 49% at $5.85 billion, connected TV reached $2.66 billion and digital $1.91 billion, while cable fell 7% and radio 13% against 2024 (AdImpact, reported by CNBC, November 17 2026).

Read that again as an operator. A dedicated measurement firm, working on the most disclosed advertising market in the world, revised its own forecast by nearly half a billion dollars inside five months. Your third-party estimate for a mid-market SaaS competitor is not more precise than that. It is the same kind of instrument pointed at a murkier target.

So use estimates as an index, never as a total. Track the direction and the size of the change between two snapshots, not the absolute figure. A competitor that moves from a modelled $40k a month to a modelled $90k a month has doubled their pressure on a channel. Whether the true numbers are $31k and $74k is a rounding argument that will burn an hour and change no decision.

## Which signals survive into a decision-grade metric set?

Six observations hold up well enough to build a threshold on. Pick three, set the thresholds, and leave the rest for the months when you have spare analyst time.

- Creative longevity: how many days a piece of creative keeps delivering. Long runs are the closest public proxy for performance you will get, and they cost nothing to count.
- Net-new creative rate: new creatives per week per competitor, and how fast the older winners get retired. This reveals production capacity and testing discipline.
- Message mix: the share of live creative in each proposition, such as price, speed, social proof or risk reversal. Shifts here usually precede a positioning move.
- Placement spread: which networks and geographies appear and disappear between snapshots. A retreat is often cheaper to detect than an entry.
- Landing page rotation: new destinations behind familiar creative. A new page with the same ad often means an offer test, not a brand campaign.
- Share of voice proxy: competitor creative volume against your own in a defined keyword or category set, used as a ratio rather than a number.

Of those six, longevity is the one to trust first. It is an observation about behaviour rather than an inference about money, and it survives every platform change we have seen since 2023. A creative running for nine weeks is telling you something a modelled spend figure never will.

## How do you run a weekly advertising intelligence loop?

Weekly is not a preference, it is the cadence where the evidence shows a difference. In Crayon's 2026 State of Competitive Intelligence, teams that share competitive intelligence weekly or faster report revenue impact at 79%, against 41% for teams on a monthly-or-slower rhythm. Only about 56% of teams hit weekly or faster, which means the advantage is available to anyone willing to hold a meeting.

A loop that works inside one working week, for a team of one to five:

- Monday, fifteen minutes: pull the snapshot and diff it against last week. Log only changes, never the full inventory.
- Tuesday, thirty minutes: write one claim per material change, in the form changed, so what, therefore.
- Thursday, fifteen minutes: take the claims to the person who owns the budget and record a decision, including decided to do nothing.
- Friday, five minutes: archive the snapshot so next week's diff has a baseline, and note what you would have missed.

The output artefact is one page. Three claims, three decisions, one line on what to watch next week. If the page grows past one screen it stops being read, and an unread report is indistinguishable from no report.

## What can advertising intelligence not tell you?

The limits are worth stating out loud, because the failure mode is a team that trusts a model more than their own account data. Four things are permanently out of reach.

- Return on ad spend for a competitor. Creative and spend signals do not carry conversion data, and anyone selling you their ROAS is inferring it.
- Audience targeting on ordinary commercial ads. Libraries expose creative, not targeting, outside political and EU disclosure regimes.
- Incrementality or true lift. That requires an experiment you control, not a view of someone else's creative.
- Private placements: direct buys, newsletter sponsorships, podcast reads and most creator deals leave no library record at all.

A competitor can look quiet in every library and still be outspending you three to one through channels that never publish an ad record. Treat absence of evidence as a coverage gap to be named in the report, not as evidence of absence.

## How AI agents change the collection layer

Collection used to consume the week: search a library, open forty ad cards, copy them into a sheet, repeat for each competitor and network. That is precisely the work an agent does well, because it is repetitive, structured and verifiable. A wired agent can query the libraries, deduplicate creative by cover asset, diff two snapshots and return the deltas with links back to the source ads.

The wiring matters more than the model. Our guide to MCP servers for ad intelligence covers how an agent reaches platform data and what a read-only scope means for a live ad account. The short version: give the agent a read path to libraries first, and treat any write path to a campaign as a separate decision with a separate approval.

The payoff is not hypothetical. In the same Crayon 2026 survey, teams running AI agents in production or pilot report revenue impact at 82%, against 42% for teams that do not, and half of teams now run agents at that stage. The gap is widest where the agent sits inside a workflow that already has an owner, which is another way of saying agents amplify a working loop and cannot substitute for a missing one.

What breaks first is unglamorous: authentication expiry, rate limits during a large pull, and schema drift when a platform renames a field. Budget for a monitor on the pipeline itself, or you will discover an empty weekly report three weeks after the credential lapsed.

## Where teams get this wrong in 2026

Four patterns account for most wasted spend, and each has a cheap correction.

- Hoarding without interpreting. If the weekly output has no claim in it, the team has built a screenshot archive with extra steps.
- Obsessing over the spend estimate. Direction and placement changes carry more decision value than a modelled total you cannot verify.
- Tracking only the three biggest rivals. The positioning move that hurts you usually starts with a smaller advertiser testing an angle nobody else will run.
- No owner and no threshold. Without a pre-agreed trigger, every finding becomes a debate, and debates resolve to whatever the loudest person already wanted.

Crayon's 2026 figures put a number on the last one. Teams that track defined KPIs report a rising competitive win rate at 66%, against 24% of teams that do not, and KPI adoption has climbed from 30% in 2022 to 60.5% now. Measurement is not paperwork. It is the difference between a programme that earns budget and one that gets cut.

## What should a small team buy first?

Buy coverage before you buy analysis, and buy one layer at a time. The tier framework in our guide to ad intelligence tools for agencies is a reasonable shopping order for a small team: free transparency archives first, one paid scraper for creative volume second, an API for anything you want to automate third. If you are still weighing a licence against building it, what an ad intelligence licence actually buys you is the unromantic version of that decision.

Start with the question you cannot answer today, not with the tool that looks most impressive. If you do not know which keywords your competitors bid on, that is a keyword intelligence problem, and finding the keywords competitors bid on in Google Ads is a cheaper first project than a platform subscription. If you cannot tell whether a rival's creative is winning or dying, you need a library and a diff, which is a week of work and no licence.

The honest sequence for a team of one: pick three competitors, pick one channel where you actually compete, run the weekly loop by hand for a month, then automate the part that hurt. Automating before the loop exists just makes a faster version of a report nobody reads.