# Study B — What contractors complain about in marketing-agency reviews (method, written before data collection) Date: 2026-10-29. Collected by AdCamp (Claude research agents), verified by a separate pass. ## Question When a small service business hires a marketing agency and it goes wrong, what went wrong? Counted from public reviews, so the categories come from what reviewers wrote, not from what AdCamp assumes. ## Sources Public review pages with dates and ratings: - Trustpilot: agencies and marketing vendors that sell to small local businesses and contractors (e.g. Scorpion, Hibu, Thryv, LocaliQ/ReachLocal, Townsquare Interactive, Web.com, Yodle/Web.com, Dex/Thryv, Service Direct, Marketing 360, Wix/Vistaprint marketing services where they sell ad management). The agent records which it used. - Better Business Bureau: complaint text for the same vendors. - Reddit: planned (first-person owner accounts in trade and small-business subreddits). Attempted 2026-10-29; reddit.com, old.reddit.com and the JSON endpoints were not reachable from the collection environment, and the search layer returned no reddit.com results. Zero rows collected; the file B-reddit.csv exists with a header only. Third-party snapshot mirrors were not used because excerpts could not be checked against the live thread. A later pass from a different environment can add this arm without changing the coding scheme. Google Business Profile reviews are not used (not fetchable). ## Sample - Target: 400 negative reviews (1–2 stars, or a BBB complaint, or a first-person negative Reddit account). Minimum accepted: 250. - Also record 100 positive reviews (4–5 stars) and tag what they praise, so the article isn't one-sided. - Reviews from 2024–2026 preferred; older ones allowed and dated. ## Complaint categories (tag up to 2 per review; add a new one only if nothing fits, and say so) - C1 contract: locked in, auto-renewed, hard or expensive to cancel - C2 opaque: no reporting, couldn't see what was done, no access to accounts - C3 ownership: didn't own the website, domain, ad account or phone number; lost them on leaving - C4 leads: few, fake, spam, or wrong-type leads - C5 overpromise: promised rankings, lead counts or ROI that didn't happen - C6 communication: unresponsive, account rep kept changing - C7 billing: overcharged, hidden fees, billed after cancellation - C8 cookie-cutter: templated site or content, nothing specific to the business - C9 spend: ad budget wasted, mismanaged, or markup on spend not disclosed - C10 ethics: fake reviews, spam tactics, misrepresentation of who they are Praise categories (positive reviews): P1 results (calls/leads), P2 communication, P3 transparency/reporting, P4 price/value, P5 speed/setup. ## Columns source (trustpilot/bbb/reddit), vendor_or_thread, url, review_date, rating (1–5 or "complaint" or "reddit"), reviewer_trade (if stated, else blank), sentiment (neg/pos), cat1, cat2, excerpt (at most 25 words, verbatim), notes. ## Rules - Excerpts are at most 25 words and verbatim, with the URL, so every row can be checked. - No names of individual reviewers. - One row per review. Skip reviews with no content ("terrible, avoid"). - Vendors are named in the data file and the method section. Whether the article names them is a separate decision for Danny. ## What changed during collection and verification (recorded 2026-10-29) - Two categories were added when reviews didn't fit the ten: C11 non-delivery (paid-for website or ads never launched or produced) and C12 broken or inaccurate deliverable (site errors, wrong services or phone on listings, botched Business Profile edits). Because C12 arrived mid-collection, an independent check found deliverable-quality complaints filed under C8/C9; all 48 C8/C9 rows were re-read and five moved to C12 (rows 28, 31, 47, 53, 220 of the original file); one unsupported second tag removed (row 242). - Two rows dropped: a pre-sales reviewer who never became a customer (rule: customers only), and a title-only review that duplicated the next row. Final file: B-trustpilot-bbb-v2.csv, 394 rows (319 negative, 75 positive). The original is kept as B-trustpilot-bbb.csv. - Trustpilot's star and sort filters refuse automated fetches, so negatives come from default-order pages 1–10 per vendor; for the two largest profiles, older negatives beyond page 10 were unreachable. BBB complaint text was reachable only for Hibu and Scorpion. - Dates are the posting date; two rows were found to carry the "date of experience" instead (same period, no effect on inclusion). - Headline shares use the seven agencies (Thryv, Hibu, Marketing 360, Townsquare Interactive, LocaliQ, Scorpion, LYFE): 260 negative reviews. Service Direct (a pay-per-call lead marketplace, 40 rows) and YP/yellowpages.com (a directory ad product owned by Thryv, 20 rows) are reported separately, since they are not agencies. - Independent verification: a fixed-rule sample of 24 rows fetched against the live pages found 24/24 excerpts verbatim, 24/24 ratings correct, 22/24 dates on the posting date, and 3 of 42 category tags disputed (7%), all in the C8/C9/C12 area fixed above. ## Findings format (written after counting) One sentence per finding with the number and denominator, e.g. "In 412 negative reviews of marketing agencies that sell to contractors, the most common complaint was being locked into a contract (38%), ahead of poor leads (31%)." Category counts are reported as share of reviews (a review with two tags counts once in each).