· 18 min read

Competitor Ad Tracking: A Weekly Log That Beats Guesswork

Most competitor ad tracking is re-browsing with a memory problem. Six fields, a four-week baseline, and a change log that separates signal from noise.

A desk with a calendar and printed performance charts used for weekly tracking

Almost everyone who “tracks competitors” is actually re-browsing them. You open the ad library, you look at a page, you notice that a brand has a lot of video running, you feel informed, and you close the tab. Next month you do it again and form a slightly different impression. Nothing is comparable, because nothing was written down in a form that can be compared.

The fix is not more discipline, and it is not a bigger spreadsheet. It is choosing the right unit, recording six fields that each stand for a decision the advertiser made, and waiting long enough to know what normal looks like before you interpret anything.

This guide is only about that layer. If you still need the mechanics of the interface, the filters and what each card field means, start with how to read competitor ads for free. If you want to query the library rather than browse it, the search grammar guide covers that. And if you are trying to work out how long a single ad has survived, the longevity guide does the date arithmetic properly. What none of those cover is the part that decides whether tracking produces knowledge or just a recurring feeling: the series.

A Snapshot Is an Opinion; a Series Is Evidence

Here is the uncomfortable property of the Meta Ad Library. It shows you the present. It does not show you a chart of how the present came to be, and outside a few jurisdictions it does not keep a history you can go back and read.

That means a single visit produces exactly one kind of output: an impression. “They are scaling video.” “They cut back.” “They are testing a new angle.” Those sentences feel like observations, but they are comparisons against a baseline you are holding in your head, and your head is a poor instrument. It remembers the last time you looked, which might have been three weeks ago, and it remembers whatever surprised you, which is not the same as what was large.

A series fixes three things at once. It replaces memory with rows, it replaces surprise with a range, and — most importantly — it forces you to define what counts as a change before you see one. That last part is where most tracking systems quietly fail, because they are built to record and not to judge.

Three Ways Tracking Fails Before You Record Anything

The identifier is not stable. If your notes say “the green one with the dog”, you have no record. Ad cards are regrouped, reordered, and re-rendered, and the same visual can be attached to several ad records across markets. The only durable key in the interface is the Library ID printed on the card. Track IDs or track nothing.

You are tracking attributes instead of decisions. Format, hook, colour, creator — these describe what an ad looks like, not what the advertiser chose. An advertiser does not decide to “run a UGC video”. They decide to keep funding an angle, to duplicate a winner into a second market, or to kill a test batch. If your fields cannot represent those choices, a year of rows will tell you that brands post videos.

You have no noise floor. Live ad counts move every week for reasons that have nothing to do with strategy: review queues clear in batches, a test batch finishes, someone pauses a duplicate. Without a measured range for normal movement, every wobble looks like a signal, and you will spend the quarter narrating noise.

The Unit Is the Portfolio, Not the Ad

Track the advertiser page as a portfolio. Not one ad, not the whole category — one page, read as a set of live bets, once a week.

It also helps to know what this layer is not. Ads are one field in a wider store-level loop that includes catalogue size, price basket, review volume and theme signature; if your questions are about the business rather than the creative, the weekly store change log is the right instrument. This one is narrower on purpose, because ad decisions turn over far faster than store decisions and need a different cadence.

Ten pages is the practical maximum for one person doing this by hand in about twenty minutes, and the composition matters more than the count. A workable watchlist of ten:

  • Three direct competitors. Same product, same buyer, same price band.
  • Three brands two steps ahead. Slightly bigger, slightly more sophisticated, still in your category. These are the ones whose decisions are copyable rather than merely impressive.
  • Two category leaders. Not to copy — to see where the ceiling is and what a mature creative pipeline looks like.
  • Two adjacent-category brands. Same buyer, different product. This is where new angles usually appear first, because they are not yet saturated inside your category.

Then apply one exclusion, and be strict about it: if a page has run the same five ads for six months, drop it. A brand that is not making ad decisions is a bad instrument for learning how decisions work. Ten pages that move beat thirty that sit still.

Six Fields, Each Encoding a Decision

Record these once per page per week. The point of each field is not the number; it is the decision that a change in the number implies.

FieldWhat you recordThe decision it stands for
Live adsCount of active ad records in your marketHow much they are willing to keep in market at once
EntriesLibrary IDs present now, absent last weekProduction rate — how fast they can ship
ExitsLibrary IDs present last week, absent nowThe kill decision, and it is the most informative one
Oldest live start dateThe earliest “Started running on” still activeWhat they are protecting
Cluster concentrationVariants in the largest cluster ÷ live adsWhether they are defending one angle or still hunting
Offer tokenDiscount depth, bundle shape, shipping thresholdPricing posture, and whether it is being tested

Live ads is capacity, not success. Forty live ads means a team that can produce and review forty at a time. It says nothing about whether any of them work.

Entries measure throughput. A page that adds eight records a week has a pipeline; a page that adds one has a founder with a camera. Neither is better — they are different operating models, and you need to know which one you are watching before you interpret anything else.

Exits are where the information is concentrated, and almost nobody logs them, because the library makes disappeared ads invisible by default. An ad that is gone is a decision someone made with money on the line. An exit wave is the closest thing to a public statement of “this did not work” that exists.

Oldest live start date tells you what is being defended. If the oldest ad has been running for 240 days while everything else turns over in three weeks, that ad is the anchor, and the rest of the portfolio is arranged around it.

Cluster concentration separates testing from milking. Nine variants on one creative out of thirty-four live ads is a brand iterating on something proven. Thirty-four ads with one variant each is a brand still searching, and the category may simply not have a settled angle yet — which is itself a finding.

Offer token is the cheapest field to record and the easiest to misread. Write it verbatim: “20% off”, “3 for $99”, “free shipping over $60”. Do not summarise it as “discounting”. The shape of the offer is the strategy; the depth is only the price of it.

Record It in Two Tables, Not One

One table cannot hold both questions you are asking. Use two, joined by Library ID.

Table one: page-week. One row per page per week. Date, page, market, live count, entries, exits, oldest live date, cluster concentration, offer token. This is the table you read for trends.

Table two: ad spine. One row per Library ID, updated rather than appended: first seen, last seen, start date, format, offer token, landing page URL. This is the table you read when you want to know whether the thing you are looking at is new or eleven weeks old in a new card.

Two rules keep the spine honest. First, expand variant groups before counting. A card that says several ads share one creative and text is a group; the group can be regrouped by the platform, which will move your numbers without the advertiser doing anything. Count the records inside the group, not the cards. Second, fix one market per watchlist. The same creative running in three countries is three records with three IDs, and mixing them inflates both entries and exits.

The mechanical part takes about two minutes per page once the URLs are saved: filter to active ads in your market, walk the first two screens, copy IDs, and fill six cells. Do it on the same weekday, at roughly the same time, because a weekly reading is a sample of a continuous process and you want the sampling error to be constant rather than random.

The Baseline: Four Weeks to Record, Twelve to Range

This is the part that most tracking sheets skip, and it is the reason people conclude nothing after two months of careful work.

Weeks 1 to 4: record, do not interpret. You have no range yet, so any conclusion you draw is a comparison against nothing. Expect this phase to feel useless. It is the price of the next phase.

Weeks 5 to 12: build the range. For each field, note the minimum, the maximum, and the median across the weeks you have. Then compute one more number, the one that does the actual work: the median absolute week-over-week change. That is your noise floor — how much this page normally moves when nothing is happening.

Week 13 onward: judge against the range, not against last week. A practical threshold that survives contact with real pages:

  • Treat a move as a change only if it exceeds twice the noise floor, and
  • it holds for two consecutive readings, or
  • a second field moves in the same direction in the same week.

Two fields agreeing is the strongest signal available for free. Live count falling while exits spike is a cut. Live count falling while exits are normal is usually your own counting error, a market filter slipping, or a page-level interruption.

Three months is not a ritual. It is roughly the shortest window that contains a full production cycle, a full promotional cycle, and at least one quiet stretch, which is what you need before “normal” means anything.

What a Change Actually Means

Every reading has at least two explanations, and the mechanical one is usually right. Run this table before you write a sentence in your notes.

What you seeRule out firstStrategic reading if the mechanical cause is ruled outHow to tell them apart
Live count drops hard, entries normalMarket filter changed, a variant group collapsed, a page-level interruptionBudget cut or a channel shiftExits concentrated on one offer or format means a cut; scattered exits mean natural expiry
Live count rises sharply, entries highOne batch published at once after a review queue clearsA launch, not a scale-upCheck offer token on the new IDs; a launch introduces a new offer or angle, a scale-up repeats the old one
Live count rises, entries lowDuplication of existing creatives into new ad sets or countriesExpansion of a proven assetCount distinct creatives, not records; duplication multiplies records while the creative count stays flat
Entries stop for three weeksProduction pause, approval queue, holidayTeam or budget freezeA freeze usually stops exits too; a pause in production with exits continuing is a shift, not a stop
Oldest live date resetsThe anchor was retired, or your filter excluded itThey killed their best-known adWatch for a similar creative relaunching within two weeks; a relaunch is a refresh, not a retreat
One cluster’s variant count grows while live count is flatRegrouping by the platformThey are iterating on a winnerRead what changed inside the variants: first frame, headline, creator, or price
Offer deepens while live count is flatA seasonal calendar you are not seeingA margin or positioning testCheck the landing page price and whether new copy mentions the new threshold
Everything doubles in one weekDuplication across many ad setsA reach test with a short leashLook again in seven days; a test loses half its records, a commitment does not

The pattern in that table is deliberate. In every row, the mechanical explanation comes first, because the library’s rendering is a more common cause of apparent change than any advertiser’s strategy. Tracking discipline is mostly the habit of asking “what else would produce this exact number” before asking “what does this mean about them”.

The Quiet Week Is Also Data

The hardest week to record is the one where nothing happened, and it is the one people stop recording. That is backwards. A page whose live count, entries and exits all sit inside the range for six consecutive weeks is telling you something specific: the operation is stable, the pipeline is predictable, and any change you see later is more likely to be real because the baseline is clean.

The opposite is also true. A page that moves every week has a noise floor so wide that only enormous changes are readable. That is not a failure of your method; it is a property of the page, and it should change how much confidence you place in any single observation about it.

There is one more use for the quiet stretch. Because you record the same weekday every week, a page that goes quiet in the same calendar slot every quarter is probably following a planning cycle rather than reacting to results. That distinction — reactive or scheduled — is worth more than any individual ad you will ever copy.

A Twelve-Week Log, With Hypothetical Numbers

The numbers below are invented to show the method. They are not a real advertiser, and no real page’s figures should be inferred from them.

WeekLiveNewExitedOldest liveTop clusterOffer
13463214d9 of 3420% off, free ship $60
23642221d9 of 36same
33358228d10 of 33same
43575235d10 of 35same
53555242d11 of 35same
63885249d11 of 38same
74196256d14 of 41same
84067263d14 of 40same
953185270d14 of 53bundle 3 for $99, new ads only
105047277d14 of 50both
1131221284d14 of 3120% off only
123353291d14 of 3320% off only

Weeks 1 to 8 are the baseline. Live count ranges 33 to 41, median 36. Entries range 4 to 9. Exits range 2 to 8. Week-over-week movement in live count is 2, 3, 2, 0, 3, 3, 1 — so the median absolute change is 2, and the threshold for a real move is 4.

Week 9 is +13. That clears the threshold twice over, entries hit 18 against a baseline maximum of 9, and the offer token on every new ID is a bundle that does not appear anywhere in weeks 1 to 8. Read mechanically, this is one batch published together with a new price architecture. Read strategically, it is a bundle test, not a scale-up — and the distinction matters, because a scale-up is worth copying and a two-week test is not.

Week 11 resolves it. Twenty-one exits against a baseline maximum of 8, with eighteen of them coming from the bundle set; the oldest live date keeps advancing, so the anchor was never touched; the surviving cluster of 14 now represents 45% of the live portfolio, up from 26% in week 1. The bundle was killed in two weeks and the proven cluster was kept. A tracking sheet would have told you that in week 11. Browsing would have told you the brand “went all-in on bundles” in week 9 and “pulled back” in week 11, which is two impressions and no knowledge.

The other lesson is about week 12. Live count moves from 31 to 33 — inside the noise floor of 2. Nothing happened. The discipline is writing “no change” and moving on, rather than inventing a reason for two ads.

A log answers what happened, not what to do about it. Turning an observed shift into something your own team can execute is a separate discipline, and the six-field format for that handoff is covered in how to write an ad creative brief. Keep the two documents apart: the log records decisions your competitors already made, the brief commits to a prediction of your own.

Three Things the Log Cannot Tell You

Why. Exits are visible; reasons are not. A creative can be paused because it lost money, because it won and got replaced by a better version, because the product sold out, or because an account-level issue paused everything. Your log records the event and never the cause. Treat every causal sentence in your notes as a hypothesis with a confidence level attached.

Spend and return. Outside the EU and UK there is no archive, no impression range, and no spend estimate — and inside them you get binned ranges, never figures. No ad platform publishes advertiser-side performance. Anything you read about a competitor’s ROAS is inference, and inference does not belong in a log that is supposed to be evidence.

Targeting. The library shows the creative and the landing page, not the audience, the placements, or the bid strategy. Two identical ads in your log may be running against completely different audiences with completely different economics. This is why the log describes the portfolio and never the campaign.

When Ten Pages Becomes a Part-Time Job

The manual method has one clean limit, and it is arithmetic. Ten pages at two minutes each is twenty minutes. Forty pages is an hour and a half, and the moment you want a per-page chart across six months, or a comparison of eight brands in one view, the spreadsheet becomes the work instead of the record.

That is the point where aggregation earns its place rather than replacing the method: the same public ad data, grouped per advertiser, with history kept so that trends are visible instead of reconstructed by hand. The store analyzer is worth pairing with it, because an ad portfolio tells you what a brand is buying and the store behind it tells you whether the traffic can convert. If you want to check whether the ads are turning into orders at all, the best-selling product research method is the closest free proxy.

Apply WINNING for 20% off TrendTrack →

Verified 2026 · 20% off Starter, Pro and Business for your first 3 months

TrendTrack promo code is WINNING - 20% off all plans for the first 3 months, applied at checkout on either billing cycle.

FAQ

How often should I check a competitor’s ads?

Once a week, on the same weekday, is the right cadence for almost every small team. Daily reading produces mostly noise, because ad counts and review queues move for mechanical reasons within a day. Weekly catches launches within seven days, which is fast enough to matter and slow enough to keep the series comparable over months.

How many competitors can one person track for free?

About ten pages, at roughly two minutes each, if you are recording six fields and maintaining an ad spine by Library ID. Beyond that the manual loop starts to eat the time you need for interpretation. Ten pages that actually change their creative are worth far more than thirty pages that have been running the same five ads since spring.

How long before I can trust a change I see?

Four weeks to record without interpreting, then eight more to establish the range and the noise floor, so roughly three months before your first confident call. Before that, you are comparing against memory rather than a baseline, and almost every movement you notice will turn out to be normal variation for that particular page.

Can I still see ads that have stopped running?

It depends on where they were shown. In the EU and UK the library keeps a window after the last impression, while campaigns shown elsewhere generally leave nothing behind once they stop. That asymmetry is the argument for your own log: outside those markets, whatever you did not record did not happen, and no tool can recover it for you.

Do I need a paid tool to track competitor ads?

No, not for reading. The free library plus a two-table sheet will carry ten pages indefinitely if you keep the cadence. A paid layer becomes justified when you need aggregation across many advertisers, long history you did not record yourself, or charts that would otherwise take an afternoon to rebuild by hand.

How do I tell whether an ad was killed or just expired?

You usually cannot tell from one reading, and you should write that uncertainty down instead of resolving it by intuition. Two things narrow it: whether the exit sits inside a wave of other exits on the same offer or format, and whether a similar creative relaunches within a fortnight, which points to a refresh rather than a retreat.

Final Word

The reason to keep a competitor ad log is not to know what your competitors are doing this week. It is to be able to say, with a straight face and a dated row behind you, what changed, when it started, and how unusual it is for that page. Six fields, two tables, one fixed weekday, and three months before the first real conclusion.

Everything else — the tools, the dashboards, the alerts — is a way of buying back the hours, not a substitute for knowing what you are counting.

    Share:
    Back to Blog

    Related Posts

    View All Posts »