Competitor ad research with an LLM:
what meal-kit brands are running

A new meal-kit or prepared-meal brand is planning its first US campaigns. Who is it competing with for attention, what do their ads promise, and how long do they run them?

The Meta ads come with their copy, headline, call to action and format. The Google ads come with their format and how many days each has been served, but not their text. The two sources answer different questions, which is why we used both.

claude-sonnet-5.5 read all 150 Meta ads in one call. For each it named the advertiser type (meal kit, grocery, diet shake, kitchen product, creator…), up to two angles (price, convenience, health, freshness, family…), the offer (percent off, free shipping, first-box deal, none…) and one short quote copied from the ad. Code checked every quote against the ad text (all 150 matched) and did all the counting. A last call turned the counts into a brief.

Most ads for "meal kit" aren't from meal kits. Of 105 unique creatives (more on that number below), only 20 came from meal-kit or prepared-meal sellers, run by 13 advertisers. The rest: 15 diet shakes and supplements, 15 kitchen products, 12 creators and media, and 40 others, from pet food to deal sites. A new brand bidding on this keyword competes with all of them for the same feed.

Meal kits lead with price. 12 of the 20 meal-kit creatives lead with a price or discount, 6 with fresh ingredients, 5 with a health goal and only 4 with saving time. 14 of 20 carry an offer: percent off (6), free shipping (4), a first-box deal (2) or dollars off (2). Diet shakes do the opposite: 10 of 15 lead with a health goal and 9 of 15 have no offer at all.

Video wins on Meta, text on Google. 50 of the 105 Meta creatives are video, 21 dynamic creative, 19 single images. "Shop now" is the call to action on 54. On Google, 85 of the 100 ads are text search ads.

On Google, winners run for months. All 100 Google ads were still running. Their median age was 155 days, 69 had run for more than 90 days, and the oldest text ad had been running since February 2023, 1,356 days. The two brands whose oldest ad dates from August 2025 had medians of 84 and 91 days; the three with ads from 2023 had 135, 182 and 198.

Stand out on time, not price. 12 of 20 meal-kit creatives lead with price and only 4 with saving time. Test: the same offer with a price headline against a time headline ("Dinner in 30 minutes, no planning needed" is the one real example), and compare cost per first order.

Try a first-box deal. Only 2 of 20 meal-kit creatives use one; percent off and free shipping are the crowded offers. Test: one creative, two offers.

Plan for long-lived search ads. 69 of 100 Google ads have run for more than 90 days. Write a few search ads you'd be happy to keep for months rather than a new set every week.

What the data can't say: spend, impressions, clicks or sales. A long-running ad is probably working, but the libraries don't prove it.

With an API key in LOVELY_API_KEY, the same run_tool and ask helpers as in the market research case study:

meta = run_tool("facebook-ads", {"urls": [{"url": "https://www.facebook.com/ads/library/?active_status=active&ad_type=all&country=US&q=meal%20kit&search_type=keyword_unordered"}], "count": 150}, 150)
google = run_tool("google-ads-transparency", {"startUrls": [{"url": f"https://adstransparency.google.com/?region=US&domain={d}"} for d in ["example-brand.com", "another-brand.com"]], "resultsLimit": 20}, 40)

ads = [{"id": i, "page": a["page_name"], "title": a["snapshot"].get("title"), "body": (a["snapshot"].get("body") or {}).get("text"), "cta": a["snapshot"].get("cta_text")} for i, a in enumerate(meta)]
# Count each (advertiser, copy) once before tagging, then:
tags = ask("Tag each ad: advertiser_type, 1-2 angles, offer, and a quote copied exactly. Return a JSON array.", ads)

Swap the keyword and the domains for your own category. To have an agent run the whole thing and come back with the brief, connect it to the MCP server.