Competitor ad analysis by market means pulling each local rival's ads from public ads libraries, labelling every ad by the tension, persona and offer it targets, clustering those labels and comparing shares one market at a time. A theme counts as whitespace only when local competitors are quiet on it and another signal, such as creator content, shows the audience cares.
Key takeaways
- Meta's Ad Library and Google's Ads Transparency Center are the two public sources most brands need, and both can be read market by market.
- Use one labelling codebook across markets so shares are comparable, but let the clusters form inside each market.
- Compare each cohort's share of its own output, not raw counts, and read volume alongside share.
- Brand absence is never whitespace on its own; without an audience-side cohort, the honest verdicts are contested, blocked or partially defensible.
- Repeat the read every quarter: SOMIN's four quarterly studies of its own competitors' paid ads covered 982 ads and showed themes appearing and fading.
Where do you find local competitors' ads?
The two public sources that matter most are Meta's Ad Library and Google's Ads Transparency Center. Both are free, both are searchable by advertiser, and both let you scope what you see to one market.
- Meta Ad Library. When Meta expanded it in 2019, the library brought in all active ads any Page is running, searchable by Page as well as by keyword, and it houses ads for seven years. For ads delivered in the EU, Meta's Digital Services Act update adds the dates each ad ran and the targeting parameters used, such as age, gender and location.
- Google Ads Transparency Center. Google describes it as a searchable hub of all ads served from verified advertisers across Search, YouTube and Display, showing which ads ran in a given region, the last date an ad ran and its format.
The first decision is the competitor list, and it should be local. In Hong Kong food delivery, SOMIN benchmarked Keeta against FoodPanda and Deliveroo. In Philippine digital banking, Tonik Bank's set was seven local challengers, from Maya and MariBank to GoTyme Bank. A single global competitor list misses the brands that set the tone in each market.
Know what the libraries do not tell you. The Tonik corpus, built from Google Ads Transparency data, held no spend or click-through data, which is why SOMIN scores creative with predicted clickability, a pre-launch estimate rather than a measured outcome.
How should you label each ad?
Label every ad against one codebook used in every market, so that a share in Manila can be compared with a share in Hong Kong. The labels describe what the ad is for, not only what it looks like.
| Label | What it captures | Example from SOMIN's reads |
|---|---|---|
| Tension | The problem the ad promises to solve | Affordable, convenient food in Hong Kong delivery |
| Persona | Who the ad is written for | The Flavourful Culinary Enthusiast (Keeta HK category) |
| Angle | The behavioural lever, with the theory behind it | Herd behaviour, loss aversion, social proof |
| Product and offer | What is sold, and the incentive attached | Credit Builder Loan, 109 of Tonik's 126 ads |
| Format and funnel stage | Search, Display or video; awareness to conversion | Google Search, Display Network and YouTube (Tonik) |
| Market and language | Where the ad ran and in which language | Country filter set before any clustering |
AI does the heavy lifting here. SOMIN's models label each ad and group similar tensions into themes with a sample size, so a theme reads as "13 ads", not as a hunch. An analyst still checks the clusters, merges near-duplicates and flags any theme that one advertiser carries alone.
How do clustering and whitespace verdicts work?
Cluster within each market, then compare three cohorts: your brand, local competitors and an audience-side cohort such as creators. A whitespace verdict needs all three.
SOMIN's rubric is strict on purpose. Brand-low alone is not whitespace; a territory earns that verdict only when brand and competitors are both low and the creator cohort, or another audience-side signal, is present. The other verdicts tell you what to do instead: contested ground needs differentiation, blocked ground is owned by someone else, and emerging-claimable ground is worth testing early.
Two cases show why the rule matters. In the Keeta HK read of 145 ads, the category's dominant tension was affordable, convenient food: FoodPanda dominated it, Deliveroo also addressed it and Keeta lagged. A GWI check then showed that Hong Kong food orderers in the above-average income bracket under-indexed on budget-driven behaviour, so the creative moved to curation and quality instead of a price fight.
In the Tonik Bank read of 126 Tonik ads against 516 from rivals, all nine tensions already carried brand or competitor presence. With no third cohort, verdicts were capped at contested, blocked or partially defensible, and the briefing's advice was to "differentiate within contested lanes" rather than hunt for empty ones.
What did SOMIN find in its own competitors' paid ads?
We run the same method on our own category. In SOMIN's audience-research workspace, four quarterly concept-whitespace studies read the paid ads of four audience-research and measurement brands that compete for the same marketing budgets. Together they cover 982 ads.
- Q1 (7 July to 6 October 2025), 279 ads. Two broad themes dominated: a digital-disconnect theme of 133 ads and an insight-deficit theme of 124. A theme of 9 ads on the trust gap in AI adoption in Thai marketing was specific to one market, the kind of signal that is easy to lose when markets are pooled.
- Q2 (7 October 2025 to 6 January 2026), 308 ads. The largest theme, on gaps in consumer insights and audience representation, ran to 193 ads, and a single global measurement company placed 120 of them. That is contested ground for everyone else.
- Q3 (7 January to 7 April 2026), 166 ads. A seasonal theme appeared: festive costs around Lunar New Year, 13 ads, 12 of them from one Southeast Asian survey platform.
- Q4 (8 April to 6 July 2026), 229 ads. A 33-ad theme warned against over-reliance on generational labels and against misreading behaviour around traditional holidays.
Two shorter trend windows inside Q4 show momentum: 19 ads in the first window (8 April to 22 May) and 63 in the second (23 May to 6 July), where a theme about rewarding everyday screen time alone accounted for 30. Reading the windows, not only the quarter, is what separates a growing theme from a one-off burst.
Because these studies hold competitors' paid ads only, they give a competitor map, not a whitespace verdict. The verdicts come when the map is set against SOMIN's own posts and the wider conversation, which run as separate 30-day whitespace studies in the same workspace.
How do you turn the verdicts into a brief?
Turn each market's strongest verdict into one lead pillar, then score and test it before you scale. A verdict without a creative plan is only a chart.
- Pick the territory with the best mix of size, verdict and permission: a real product or heritage asset that makes the claim true.
- Write concepts for each funnel stage and score them before spend; SOMIN's paid media agent predicts click-through from what already wins in your category.
- Re-test dropped ideas that scored well. Tonik's ₱20K medical-needs micro-loan angle scored 0.514 predicted clickability, the highest in its corpus, yet ran in only 2 ads before it was discontinued.
- Hand the verdict map to strategy so that pillar, persona and moments line up; the SOMIN strategy agent turns it into a client-ready brief.
What mistakes distort a market-by-market read?
Most errors come from mixing markets or mixing cohorts. These are the ones we check for first, and our guide to share of voice by market covers the measurement side in more detail.
- Pooling markets. One loud market drowns the rest. In the Defender read, US engagement was 1,619,446 against 40,756 in Japan.
- Comparing raw counts. A creator panel and a competitor cohort can differ in size many times over, so compare each cohort's share of its own output.
- Reading share without volume. A falling share can hide rising volume; check both before calling a decline.
- Trusting thin samples. The Defender US creator panel had only 5 Adventurer-tagged posts in 12 months, below SOMIN's clustering threshold, so US verdicts fell back to brand and competitor cohorts at moderate confidence.
- Mistaking one advertiser for a trend. A theme carried by one brand may be that brand's campaign, not a category signal.
FAQ
Can I analyse competitors' ads from the Meta Ad Library and Google Ads Transparency Center?
Yes. Both are public and free to search, and they exist to show who is advertising and what they run. Reading them is standard competitive research; follow each platform's terms if you collect ads at scale.
Do ads libraries show competitors' spend or results?
Treat them as a map of what rivals say, not how well it worked. The Google corpus behind SOMIN's Tonik read held no spend or click-through data, so performance came from predicted scores and later tests.
How many ads do I need per market?
There is no universal number, but every theme should carry its sample size and a confidence flag. When a cohort falls below the clustering threshold, as Defender's US creator panel did with 5 posts, base the verdict on the other cohorts and say so.
How often should competitor ad analysis be repeated?
Quarterly is a sensible rhythm, with shorter trend windows inside each quarter. SOMIN's own studies ran four quarters plus two shorter trend windows to separate lasting themes from bursts.
Pick your three priority markets, list the local competitors in each and pull 90 days of their ads from both libraries. Label them against one codebook, cluster inside each market and write down a verdict per theme before anyone writes a headline.
