How NiceToMeet protected a 45+ audience from romance scammers and turned 5,800 monthly comments into $78,400 of new memberships

NiceToMeet sells one promise above everything else: that a stranger you meet at one of their dinners will be exactly who they say they are. The product is small-group meetups for adults 45 and over. The matching quiz is short. The venues are vetted. The whole experience is designed to feel like meeting a friend of a friend, not a Tinder date.

That promise made the comment section a problem. Every paid Instagram and TikTok creative drove genuine prospects to the brand: "How does this work?", "Where are you in Atlanta?", "Cost?". It also drove a second crowd that the team had not budgeted for. Romance scammers, lonely-hearts impersonators, and affiliate-link bots aim for exactly the same demographic. Adults 45 and over, comfortable enough to comment publicly on a friendship ad, are the highest-value cohort for these scams in the entire social ecosystem.

Worse, the comment section was the first place a skeptical prospect went to check whether the brand was real. A scrolling 52-year-old who had finally worked up the courage to consider attending would scroll past the comments before they ever clicked the matching quiz. If the third comment they saw was @matchmaker_anna_xo offering "private connections" or @lonely_dave_72 begging for WhatsApp messages, the brand promise of safety was already broken before any conversation began.

The Community Experience team was triaging ~30 "is this a scam?" support tickets per week, all from people who had been spoken to in DM by an account they assumed was a NiceToMeet team member. None of those accounts were the brand. All of them had replied first.

FeedGuardians ran a 7-day read-only audit across @nicetomeet on Instagram, the TikTok account, and the active paid creative. Across the audit window, the system logged 5,820 comments and classified each one by intent and risk profile.

The audit reframed the team's mental model. The brand had been thinking of comments as "questions plus some spam to clean up". The data showed that predator-pattern comments were structurally larger than the team had accounted for, and that they were arriving faster than the brand voice could.

The most damaging pattern was not volume. It was timing. The first comment under a high-performing creative usually arrived within 4 minutes. Half of those first comments came from new fresh-creation accounts following a recognizable pattern. The longer one of those accounts sat at the top of a comment thread, the more skeptical comments ("Is this a scam?", "I don't trust this") followed beneath it. A single visible predator comment in the first hour of a viral post measurably reduced quiz-completion rate on the linked landing page by 18 to 24 percentage points.

The rollout had three parts.

1) Predator filter. A pattern-matching ruleset was trained on 31 distinct romance-scam comment shapes and on 18 fresh-creation account heuristics specific to the friendship-product space. Any comment matching the filter was hidden within 30 seconds of posting. New variants surfaced weekly were added to the rule set.

2) Genuine-inquiry auto-reply. The AI reply was trained on the NiceToMeet membership FAQ, current city coverage, the matching-quiz copy, and recent positive Trustpilot reviews. When a comment was classified as a genuine inquiry, the system posted a localized public reply ("We have weekly dinners in Atlanta. Take the 2-min quiz at nicetomeet.com/quiz to see if it's a fit") and triggered a DM with a pre-filled inquiry form. The whole loop completed in under 60 seconds.

3) Trust-signal moderation and member-safety escalation. Trolls and age-mocking comments were hidden so the visible comment section reinforced rather than undermined the brand promise. Any comment that named a specific member, included contact information, or referenced a member's real-life identity was routed to a Slack channel monitored by the Community Experience team. Roughly 4% of comments needed human attention; the AI handled the other 96%.

Week 1: predator visibility on commented posts dropped from a 22% share to under 3%. The "first comment" position on new creative was almost always a NiceToMeet auto-reply rather than a scammer.

Week 3: monthly qualified membership inquiries climbed from a baseline of 110 to 560. A 5.1× lift. Average comment-to-DM response time dropped from ~3 hours to 38 seconds. The matching-quiz completion rate on social-driven traffic recovered the 18 to 24 point penalty observed during the audit.

Week 6: city-aware auto-replies launched. Atlanta prospects got Atlanta replies, Austin prospects got Austin replies. DM-to-membership conversion lifted another 14% on top of the inquiry volume gain.

Week 8 (end of pilot): "is this a scam?" support tickets dropped 91% versus the pre-pilot baseline. The Community Experience team's manual-moderation hours fell from ~22 hours/week to ~5 hours/week. That capacity was redirected to facilitating post-event "stay in touch" connections, which is the work the brand actually wanted that team doing.

In the pilot quarter, the comment-to-DM funnel produced $78,400 per month in new attributable membership revenue against a baseline of roughly $5,900. Net incremental: ~$72,500 per month run-rate, or ~$870k annualized at flat baseline.

The FeedGuardians subscription was $89 per month for the connected accounts. The harder-to-quantify second outcome was the brand-promise outcome. NiceToMeet's Trustpilot score has held at 4.9 since deployment, and the share of new members citing "I read the comments first and they felt safe" in the post-event survey rose from 11% to 34% over the pilot.