Why Experienced Marketers Overtrust AI: An Expert's View
Featuring insights from Igor Ivitskiy, PhD founder and Chief Scientist of Doctor Ads, creator of the Profit Forensics methodology, and a 19-year paid-search (PPC) practitioner who has managed over $770M in ad spend.
Many AI-driven platforms act as both the advisor and the judge. They recommend what to do, decide how success is measured, and then report back that they succeeded.
That's the uncomfortable premise behind this conversation. We reached out to Igor Ivitskiy, a mathematical modeler who spends his career auditing how automated ad platforms actually perform against the goals advertisers set them to understand something we think about constantly on this site: why do smart, experienced people trust systems more than the systems have earned?
His answers, lightly edited for length, map directly onto three of the most consequential cognitive biases in behavioral science: Automation Bias, Authority Bias, and Anchoring. Here's what he found and what he does about it.
The Trust Gap: Why the Confidence Doesn't Match the Performance
Why do experienced marketers and executives tend to overtrust AI-generated recommendations, even when the underlying performance doesn't justify that confidence?
"Because the same system that makes the recommendation also grades its own results, and the interface is built to display confidence, not misses. A target you typed in, a green 'Excellent' label, a confident projection, all read as a verdict, and the confidence transfers from the badge to the decision.The badges rarely earn it. 'Excellent' ad strength beats 'Average' on click-through only about 55% of the time, a coin flip dressed as a grade. And across 1,950 Target-CPA campaigns and roughly $42M in spend, only 34% landed within 10% of the target the advertiser set. The platform almost never surfaces that miss as a miss.Experience makes it worse, not better: senior people have learned that delegation works, so they extend to the machine the trust they built for capable colleagues."
This is a specific, measurable version of a pattern behavioral scientists have studied for decades, a confident interface produces confident trust, independent of whether that confidence is earned. The label "Excellent" isn't a finding. It's a design choice. And Ivitskiy's numbers show it performs barely better than chance at predicting what it claims to predict.
The Two Biases Doing the Real Damage
Which cognitive biases are most responsible for this overreliance on AI?
"Automation bias does the heavy lifting. The machine's suggestion becomes the default, so agreement is free and disagreement is the move that has to justify itself. That asymmetry is the whole problem.Authority bias adds the halo, because the advice arrives wearing the platform's brand and gets read as expert counsel rather than a vendor default. The platform's own optimization score is the clearest case: I compared an account sitting at a 99% score while leaking about 19% of its budget on searches that never converted, with a 49%-score account in the same niche that was the leanest of the set. The score tracked alignment with the platform's advice, not commercial performance.Anchoring closes the loop: once you set a target, you grade outcomes against that promise rather than your own economics even though only about 27% of Target-ROAS campaigns reach or beat the number they were set."
Automation Bias is the tendency to favor decisions made by automated systems, even when contradicting evidence is available the default becomes "trust the machine" rather than "verify the machine." Authority Bias is the tendency to attribute greater accuracy to the opinion of a perceived authority figure in this case, the platform itself, dressed in the visual language of expertise. And Anchoring, a bias you'll recognize from elsewhere on this site means the first number presented (the target) becomes the reference point against which everything else gets judged, rather than the number that actually matters (real profit).
The point worth sitting with in Ivitskiy's example: a 99%-scored account was worse at the thing that matters than a 49%-scored one. The score wasn't measuring performance. It was measuring obedience to the platform's own recommendations and being graded well for it.
What Actually Works: Precommitment
What practical habits help leaders use AI as a decision-support tool without becoming overly dependent on it?
"Precommit before you see the recommendation. Write down four things: the business metric that actually matters, today's baseline, the date you will review it, and the result that would make you stop. Once the AI's forecast is on the screen, it becomes the anchor, so the decision has to be framed before it appears.Then hold one metric the system does not optimize, usually profit, and let it be brilliant only inside an envelope a human owns.Put plainly: AI is decision-support only when it is allowed to lose the argument."
This is a structural fix, not a willpower fix and it's worth noticing the parallel to something we've argued elsewhere on this site about Confirmation Bias in customer research: the fix isn't "try to be less biased." It's building a process where the bias has less room to operate.
In plain terms, what Ivitskiy is describing is this: before you look at what the AI recommends, write down the real business result that would define success for you, the actual profit, the actual margin, the actual outcome your business cares about. Then judge the AI against that number, not the number the platform hands you. The platform can only optimize for what it can see (clicks, conversions, its own scores). Your real goal usually includes things it can't see. Committing to your own success metric before the machine's number appears means the machine's number can't quietly become the definition of "good."
Writing down your real success metric before the AI shows you a number is functionally identical to designing a research question that could disprove your hypothesis. Both work by removing the opportunity for the bias to anchor first.
The Biggest Misconception
What's the biggest misconception organizations currently have about AI-driven decision-making?
"That AI optimizes your goal. It optimizes the proxy it was handed, and the proxy drifts from the goal the moment margin, lead quality, or fraud enter the picture.It's also easy to mistake automation for optimization: in a typical account, about 62% of conversions come from the top 1% of search terms, while the system quietly funds a long tail of waste that never reaches the summary screen. Most organizations audit the outputs and no one audits the objective.The failure that should worry a board is not a visible error, it is a system executing flawlessly, at full budget, toward a goal that was set five degrees wrong."
This closing point is the one worth sitting with the longest. A system that fails visibly gets caught. A system that succeeds perfectly at the wrong objective often doesn't because everything about it, from the dashboard to the quarterly report, is telling you it's working.
The Pattern Behind the Pattern
What makes this conversation valuable isn't that AI is untrustworthy, it's that trust, in these systems, is being generated by design choices (badges, scores, confident language) that have nothing to do with the system's actual track record. The fix Ivitskiy describes, deciding what success looks like for your business before you ever see the machine's number is a direct, practical answer to Anchoring, Automation Bias, and Authority Bias operating together.
It's a genuinely useful habit for anyone making decisions alongside automated systems, in advertising or otherwise: decide what would make you stop, before the system tells you why you shouldn't.
Behavioral basis: On Automation Bias — Skitka, L. J., Mosier, K. L., & Burdick, M. (1999). Does automation bias decision-making? International Journal of Human-Computer Studies, 51(5), 991-1006. On Authority Bias — Milgram, S. (1963). Behavioral study of obedience. Journal of Abnormal and Social Psychology, 67(4), 371-378. Cialdini, R. B. (1984). Influence: The Psychology of Persuasion. On Anchoring — Tversky, A. & Kahneman, D. (1974). Judgment under uncertainty: Heuristics and biases. Science, 185(4157), 1124-1131. Industry data cited by Ivitskiy is drawn from his firm's own audit work across approximately 1,950 Target-CPA campaigns and $42M in advertising spend.
Igor Ivitskiy, PhD, is the founder and Chief Scientist of Doctor Ads, creator of the Profit Forensics methodology, and was ranked #6 in the Top 50 Most Influential PPC Experts of 2026 by The PPC Survey, an independent industry research project regularly cited by Search Engine Land and other digital marketing publications. Learn more at thedoctorads.com or connect on LinkedIn.
Read more on the biases discussed: Anchoring Effect →
