Return on ad spend is a ratio, and ratios break from either side. Teams instinctively audit the numerator, revenue, and the obvious inputs, creative, targeting, landing pages. The quieter failure sits in the denominator's shadow: a surge of bot traffic inflating sessions and clicks while adding no buyers at all. The arithmetic is unforgiving. If automated traffic adds a third more visits at zero conversion, the measured conversion rate falls by a quarter, ROAS follows it down, and every dashboard in the building reports a marketing problem. Then the second order damage starts, the one that costs real money. Ad platforms optimise toward the behaviour they observe. Bot sessions that land, bounce in patterned ways, or worse, generate fake leads and add to carts, teach the platform's models that this is what an interested audience looks like. The platform then buys more of that audience. You are no longer just diluted. You are training your own media buying to find bots.
The shape of a bot driven ROAS drop
Bot dilution has a fingerprint creative fatigue does not. It arrives suddenly rather than as a slow fade. It concentrates: one campaign, one geography, one placement or publisher segment carries most of the damage while sister campaigns hold steady. Session quality collapses in specific ways, near zero durations, pages per session pinned at one, device and browser combinations your real customers rarely use, and conversion falling hardest exactly where the session count rose most. Compare the suspect segment's behaviour against your known human baseline before touching a single ad. If the segment's sessions behave like no customers you have ever had, they are probably not customers.
Check the timing against the open internet, too. Scraping and click fraud operations ramp against retail calendars, big sale events, and product launches, because that is when budgets and cover traffic peak. A ROAS drop that starts on the first day of your biggest promotion is more likely to be an infestation than a coincidence.
Related reading on this site: Home and Application Fraud Costs Less to Catch and Pays Worse Than Claims Fraud cover the neighbouring problems in more depth.
Cleaning the signal is the repair, pausing is not
Pausing the affected campaign stops the spend and teaches you nothing, and the bots find the next campaign when it resumes. The repair is exclusion and verification: block or challenge the traffic at the edge by bot score and behaviour, exclude the poisoned placements and geographies at the platform, and add verification steps where fake conversions enter, lead forms and account creation above all, because a fake lead costs sales time on top of media money. Then accept a measurement consequence most teams resist: the historical data from the infested period is not a baseline. Conversion benchmarks, audience models, and lookalikes built over bot polluted windows carry the pollution forward, and they should be rebuilt or at least re windowed.
Report ROAS with a traffic quality line under it
The standing fix is a habit, not a tool purchase: every ROAS report carries a traffic quality companion, the share of sessions that are verified bots, scored as likely automated, or behaviourally anomalous, by campaign. When ROAS moves and the quality line moved first, the meeting is short. When ROAS moves and traffic quality is flat, you have a real marketing problem and the creative audit can start. Most teams have the argument in the wrong order for years, debating ads while the denominator quietly fills with machines.
Frequently asked questions
How does bot traffic lower ROAS without changing the ads?
Bots inflate the visit and click counts that conversion rates and ROAS are calculated from, while adding no revenue. The measured efficiency of the same ads falls, and ad platforms, optimising on the polluted behaviour, shift spend toward audiences resembling the bots.
What are the signs a ROAS drop is bot driven?
A sudden rather than gradual fall, concentration in one campaign, geography, or placement, collapsing session quality such as near zero durations and single page visits, and conversion falling most where session counts rose most. Creative fatigue fades slowly and evenly by comparison.
Why is pausing the campaign not a real fix?
Pausing stops spend but leaves the traffic source, the placements, and the polluted audience models untouched. When the campaign resumes, the same automated traffic returns. Exclusion, edge bot controls, and verification on lead and signup forms address the cause.
What should be rebuilt after a bot infestation of ad data?
Audience models, lookalikes, and conversion benchmarks built during the polluted period, since they encode bot behaviour as if it were customer behaviour. At minimum, exclude the infested window when those assets are next trained or calculated.