A $100 million-per-year government services company wanted to improve their Facebook ad ROAS. They didn't change their creative. They didn't change their targeting. They didn't change their budget. They turned on one AI agent to reply to ad comments, and their ROAS went from 1.4x to 1.7x.
That's a 21% relative improvement from a single change. The CEO attributes it directly to automated comment replies, calling it the only significant variable they altered during that period.
We pulled their full dataset to understand what happened.
A $100M/year gov-tech company improved ad ROAS from 1.4x to 1.7x by deploying one AI agent to reply to every ad comment. Over 4 months, the agent posted 33K replies containing 26K shortlinks that generated 47K clicks to their product pages.
The Setup
This company runs a single, high-volume Facebook page. One page, one AI agent, fully automated. Every comment gets a reply. No human review queue, no approval steps, no manual oversight.
Their agent configuration is straightforward:
| Setting | Value |
|---|---|
| Publish mode | Fully automatic |
| Reply coverage | Every post (ads + organic) |
| Link insertion | When relevant (AI decides) |
| Moderation | On, every post |
| Auto-like comments | Yes |
| Respond to hidden comments | Yes |
| Reply tone | Warm |
| Response timing | Natural (human-like delays) |
The notable choice here is respond to hidden comments. Even comments flagged for moderation get a reply before they're hidden. The agent replies first, then the moderation step runs. This means the commenter still sees a response even if their comment gets hidden from public view.
The Numbers
Here's the all-time summary since the agent went live in October 2025:
| Metric | Value |
|---|---|
| Incoming comments | 34,911 |
| AI replies posted | 33,483 |
| Comments moderated (hidden) | 15,616 |
| Shortlinks created | 26,502 |
| Shortlink clicks | 47,020 |
| Posts tracked | 424 |
| Reply rate | 96% |
| Clicks per shortlink | 1.77 |
The standout number is the 96% reply rate. Nearly every incoming comment gets a response. Compare this to the multi-page approach in our previous case study, where a portfolio of 31 pages averaged a 41% reply rate with selective, high-confidence replies. This company took the opposite approach: reply to everything, let volume do the work.
Growth Curve: From 4K to 10K Comments per Month
Comment volume more than doubled between October and January, driven by increased ad spend. The AI reply volume tracks almost perfectly — except for a brief dip in November where the agent was being reconfigured (reply rate dropped to ~68% for two weeks before returning to 99%+).
By January 2026, the agent was processing 10,000+ comments per month from a single page. At that volume, manual replies would require a dedicated team. The agent handles it automatically.
Every Reply Link Gets Clicked (Almost Twice)
The headline metric here is the click multiplier: across the entire dataset, 26,502 shortlinks generated 47,020 clicks — an average of 1.77 clicks per link. People don't just see the links in AI replies, they actually click them. And they click them more than once.
The chart also reveals how tightly clicks track with shortlink volume. When the agent was being reconfigured in November (only 872 shortlinks created), click volume stayed elevated at 3,732 — October's links were still generating traffic. When volume surged in January with 9,791 new shortlinks, clicks jumped to nearly 20K.
How link age affects click volume:
| Cohort (Month Created) | Shortlinks | Total Clicks | Clicks per Link |
|---|---|---|---|
| Oct '25 | 3,723 | 5,735 | 1.54 |
| Nov '25 | 872 | 2,092 | 2.40 |
| Dec '25 | 2,399 | 8,129 | 3.39 |
| Jan '26 | 9,791 | 20,014 | 2.04 |
| Feb '26 | 9,717 | 11,050 | 1.14 |
Older cohorts accumulate more clicks over time — December's links lead at 3.39 clicks each after 3+ months of exposure. February's cohort is at 1.14, but it's only two weeks old. The longer a reply link stays visible in a comment thread, the more clicks it collects.
The 96% Reply Rate Strategy
Most companies using AI comment replies take a selective approach: set a high confidence threshold, reply only when the AI is confident, skip the rest. This company does the opposite.
96 out of every 100 comments get an AI reply. The remaining 4% are edge cases — comments in unsupported languages, empty comments, or processing errors.
They also moderate aggressively: 45% of all comments get hidden. But here's the key — these aren't mutually exclusive. A comment can get both a reply and get hidden. The agent replies first, then moderation runs. So even a spam comment might get a helpful response (visible to the commenter) before being hidden from public view.
This "reply to everything, then clean up" approach has a clear rationale:
- Every reply is a potential click. By replying to 96% instead of 41%, they're creating ~2.3x more shortlinks per comment than a selective approach would.
- Hidden comments still get value. The commenter sees the reply even if others don't. That's still a potential click.
- Volume compensates for precision. Their clicks-per-shortlink (1.77) is lower than the 2.53 average in our previous case study. But with 2.3x more shortlinks per comment, total click volume is higher per comment processed.
The ROAS Impact
Before turning on the AI agent in October 2025, this company's Facebook ad ROAS was 1.4x — $1.40 returned for every $1 spent on ads. After four months of automated comment replies, their ROAS rose to 1.7x.
That's a 21% relative improvement in ad return, and the CEO says comment automation was the only significant change to their ad operations during that period.
Why would comment replies affect ad ROAS? Three reasons:
-
Higher engagement signals. Facebook's algorithm rewards posts with active comment sections. When an AI replies to 96% of comments, the post's engagement score rises, which can lower CPM and improve delivery.
-
Direct traffic from shortlinks. 47,020 clicks to product pages that didn't come through the ad click itself. These show up as website traffic attributable to the ad campaign (via UTM tracking), improving the measured return.
-
Social proof effect. A comment section full of helpful, link-containing replies looks different from one full of unanswered questions or spam. New viewers who see the ad are more likely to engage — and click — when the comment section signals trust.
Is it possible other factors contributed? Of course. Ad performance is multivariate. But the correlation is strong: ROAS was flat at 1.4x for the period before, started climbing within weeks of activating the agent, and has held at 1.7x since.
The ROI Math
Total spend over 4 months:
- Subscription: $450/month x 4 months = $1,800
What they got:
- 47,020 clicks to product pages
- Cost per click: $0.038
For context, Facebook's average cost-per-click ranges from $0.50 to $2.00+. This company is getting supplemental clicks at roughly 1/13th to 1/50th the cost of paid ads.
But the real ROI story is the ROAS lift. On a $100M/year business where a meaningful portion comes through paid ads, a 0.3-point ROAS improvement dwarfs the $1,800 subscription cost. The shortlink clicks are almost a bonus on top of the algorithmic and social proof benefits.
Want results like this?
See what one AI agent can do for your comment sections. No credit card required.
How This Compares
We've now published two case studies with very different approaches to the same strategy:
| Previous study | This study | |
|---|---|---|
| Scale | 31 pages, 26 agents | 1 page, 1 agent |
| Reply rate | 41% (selective) | 96% (everything) |
| Confidence threshold | 0.95 (strict, hybrid mode) | Fully automatic, no threshold |
| Clicks per shortlink | 2.53 | 1.77 |
| Moderation rate | 33% | 45% |
| Total clicks | 109K | 47K |
| Headline metric | Click volume | ROAS lift |
Both approaches work. The selective approach produces higher-quality replies and more clicks per link. The everything-gets-a-reply approach produces more total shortlinks and contributes to ROAS through volume and engagement signals.
The right strategy depends on your goals. If you're optimizing for click quality and brand voice, go selective. If you're optimizing for ad performance and engagement metrics, go aggressive.
FAQ
Can one AI agent really affect ROAS?
ROAS is influenced by many factors, and we can't isolate comment automation as the sole cause with scientific certainty. What we can say: the CEO of this company attributes the improvement directly to ReplyZen, the 1.4x → 1.7x shift coincided with the agent going live, and no other significant changes were made to their ad operations. The mechanism is plausible: more engagement → better delivery → lower CPM → higher return.
Why reply to everything instead of being selective?
For a high-volume single page, the math favors volume. At a 1.77 clicks-per-shortlink rate, every reply with a link generates nearly two clicks on average. If you skip 60% of comments (like a 41% reply rate), you're leaving ~60% of potential shortlinks — and their future clicks — on the table.
Does replying to hidden comments make sense?
It depends on your strategy. This company has it enabled because the commenter still sees the reply in their notifications, even if the comment is hidden from public view. That's still a potential click. For brand reputation, you might prefer to skip hidden comments entirely.
What model are they using for replies?
They use a cost-optimized model (Gemini Flash) that balances quality with cost. At 33,000+ replies over 4 months, using a premium model would significantly increase costs without proportional improvement in click-through rates.
Does this work for smaller ad spenders?
The ROAS mechanism applies at any scale — engagement signals and social proof work the same way whether you get 100 or 10,000 comments per month. The shortlink click benefits scale linearly with volume. Even at lower volumes, the subscription cost is low enough that a handful of additional conversions covers it.
Data covers October 2025 through mid-February 2026. February figures are partial-month. All data is anonymized; no company names, page URLs, or team member names are disclosed. ROAS figures are as reported by the client and have not been independently audited.
