MCP · Guide

Ad research with SponsorGap in Claude

Connect SponsorGap to Claude and ask for ad copy or a full creative brief built on real sponsorship ads: what competitors actually ran in newsletters, podcasts and YouTube, word for word, with where and when.

Two questions this guide walks through:

Claude finds the ads, reads them, spots the patterns and writes new copy. SponsorGap supplies the evidence, so the copy builds on what brands paid for, not on guesses.


Setup (2 minutes)

  1. Get your API key in Settings → API Access (Professional or Advanced plan).
  2. Claude.ai: Settings → Connectors → add SponsorGap with the MCP URL, then enter your API key. Claude Desktop: add the server to your config and restart. Step-by-step for each app: sponsorgap.com/mcp.
  3. In a new chat, open the tools menu and check that SponsorGap is switched on.
Connected before 17 November 2026? Disconnect and reconnect once (or restart Claude Desktop). Claude keeps the tool list from the last connect, so search_ads and the Ad research prompt only show up after a reconnect.

The tools Claude uses

You don't call the tools yourself. Claude picks them based on your question. Knowing them helps you ask better questions.

Tool or promptWhat it returnsGood for
search_adsAds across all brands that match any of your keywords: ad copy, where and when each ran, audience, promo code, offer, angle and hook, and whether the brand rebooked"How do AI tools pitch in newsletters?"
get_brand_adsOne brand's recent ads, with a summary of its usual offers and angles"What is Jasper running right now?"
find_competitor_sponsorsBrands sponsoring in your spaceBuilding the competitor list
find_brand_appearancesEvery newsletter, podcast and channel a brand sponsoredWhere to place your ad
get_industriesThe industry list with sponsor countsNarrowing a search
Ad research (prompt)The 7-step research brief, filled in with your product, market and problemCreative briefs with a test plan

Rebooked is the signal to watch. A rebooked ad means the brand paid the same newsletter or channel again on another day. There is no click data, so a rebooking is the best sign that an ad works. Claude weighs those ads more.


Example 1: ad copy for an AI company

The short version works, but Claude writes much better copy when it knows what you sell and who reads the ad. Add those two lines:

I run an AI meeting-notes tool for sales teams ($29/month, 14-day free trial). Look at the ads AI companies ran in newsletters and on YouTube in the last 90 days, tell me how they pitch, and write 5 newsletter ads for us based on what works. Weigh rebooked ads more.

What Claude does

  1. Calls search_ads with keywords like AI, AI assistant, AI agent, meeting notes, transcription, copilot over the last 90 days, filtered by angle or offer when useful.
  2. Calls get_brand_ads for the 2–3 closest competitors to read their full pitch.
  3. Groups the openings, promises, offers and calls to action, and notes which brands rebooked.
  4. Writes your ads in each pattern, using your own claims, and lists the source ads with links.

What the evidence looks like

A check on 17 November 2026 for Higgsfield, an AI media generator, returned 8 ads in 6 channels over 180 days. The pattern is clear: YouTube host reads in "make money with AI" videos, a creator-branded link or code ("EXTRA 20% OFF with code TILBURY"), and a repeat booking with Nick Saraev in August and October. Claude turns findings like these into rules for your copy: lead with the outcome, give the host a personal line, attach a code you can track.

Sample output (shows the format, not real results)

Your reps stop typing. Your CRM fills itself.

Every call is transcribed, summed up and logged to Salesforce before the rep hangs up. Teams get back about 5 hours a week (needs proof).

Try it free for 14 days →

Pattern: time saved + concrete outcome. Learns from: 3 rebooked AI-assistant ads in sales newsletters.

Tip: ask for "one ad per hook pattern" to get variety, then "rewrite the best two for a podcast host read".


Example 2: 12 concepts with a test plan

For a full creative brief, use the Ad research prompt. It hands Claude a fixed 7-step method, so the result has the same structure every time.

Option A: the prompt menu

Click + in the chat box → SponsorGap → Ad research: competitor ads → 12 concepts, and fill in:

FieldExample
Product (required)Fieldwork 3-layer waterproof hiking jacket, $229
Market (required)Outdoor apparel, US
Buyer problem (required)Jackets that wet out after an hour of rain
Proof (optional)4.8 stars from 1,200 reviews; 20,000 mm waterproof rating; lifetime repair

Option B: plain words

Type the request directly. Claude uses the same tools, but tell it your product and proof, or it will mark claims as "needs proof":

Research competitor sponsorship ads for waterproof hiking jackets in the US outdoor market and write 12 concepts with a test plan. Our product: […]. Our proof: […].

The 7 steps Claude follows

  1. Define the search: 6–12 short keywords (rain jacket, waterproof, hiking, trail, backpacking, Gore-Tex) plus brands like Patagonia, Arc'teryx and REI.
  2. Collect the ads: search_ads over 365 days, broader words if fewer than 15 ads come back, then get_brand_ads for the top 2–3 competitors.
  3. Remove the noise: merge repeats and drop ads that only share a word. Claude reports ads found, advertisers and relevant ads.
  4. Map the hooks: 4–8 hook patterns with a fill-in template each, plus counts of angles and offers. Rebooked ads count more.
  5. Find the gaps: 3–6 claims buyers care about that competitors under-use, each marked "backed by my proof" or "needs proof".
  6. Draft 12 concepts: title, angle, format, hook, visual, body, proof, CTA and the source ads it learns from. No copied lines.
  7. Plan the tests: a hypothesis, a control and one deciding metric per concept.

You get a closing table of the 12 concepts and a list of source ads with links. Want it saved and exportable? Run the same brief in the Ad Library Researcher, which exports Markdown and CSV.


Prompts that work well

Copy, fill in the brackets, paste.

How do [AI writing tools] pitch in newsletters? Use the SponsorGap ad library for the last 90 days, group the ads by hook and offer, and show which brands rebooked.
Based on SponsorGap ads for [category], write 5 ads for [product], one per hook pattern. Use only these claims: [proof]. List the source ads.
What ads is [brand] running right now? Summarise its pitch, then write 3 ads that beat it for [our product].
Rewrite the best two concepts as a 60-second podcast host read and a 3-line newsletter classified.
Which newsletters did [competitor] book more than once? Which of those fit [our product]?

Tips


Limits and troubleshooting

ProblemWhyFix
Claude doesn't use search_ads or show the Ad research promptOld tool list from before 17 November 2026Disconnect and reconnect SponsorGap (Claude Desktop: restart)
Few or no ads foundKeywords too long or too abstractAsk for shorter, concrete words and competitor names
Some ads only have a summaryWord-for-word copy is stored for ads found since 7 November 2026Claude uses the summary and quote; the issue link opens the full ad
Copy says "needs proof"No proof was givenAdd reviews, numbers or guarantees you can back up
No click or conversion dataSponsorGap sees placements, not resultsUse rebookings as the signal and run the test plan

Try it with your own product

Available on Professional and Advanced plans.

Connect SponsorGap to Claude →