TL;DR
- StackAdapt published "The AI Delegation Gap" on 19 August 2026, based on a NewtonX survey of 500 marketing and advertising professionals fielded 19 May to 8 June 2026 plus 187 StackAdapt clients.
- Only 6% of marketers say they act on in-platform AI recommendations almost always, while 91% say their organisation uses AI tools and 88% report some AI-driven performance improvement.
- Comfort drops along the autonomy ladder: 90% accept AI recommending an action, 89% AI preparing one for approval, 78% AI acting within human-set rules, but only 50% accept full autonomy even after performance is proven.
- Reasons for ignoring recommendations: 42% generic or irrelevant, 22% not aligned with strategy, 17% no explanation, 12% timed to drive spend, 6% too many to evaluate, 1% other.
- What makes marketers act: 33% a clear explanation, 31% a clear tie to a KPI. Across relevance, timing, clarity, manageability and ease of action, no dimension scored above 40% positive. No Thailand breakdown was published.
StackAdapt published a study called "The AI Delegation Gap" on 19 August 2026 reporting that only 6% of marketers act on the AI recommendations inside their advertising platforms almost always. The figure comes from a NewtonX survey of 500 marketing and advertising professionals fielded between 19 May and 8 June 2026, run alongside a parallel survey of 187 StackAdapt clients. The same respondents describe AI as already embedded in their daily work, with 91% saying their organisation uses AI tools in marketing or advertising.
StackAdapt fielded the research between 19 May and 8 June 2026
The study StackAdapt released on 19 August 2026 was conducted by NewtonX and covers 500 marketing and advertising professionals surveyed between 19 May and 8 June 2026, plus a second and separate survey of 187 StackAdapt clients. PPC Land reported the release and led on the same number that anchors the study, the 6% of marketers who say they almost always act on the recommendations their ad platforms serve them.
Adoption, as StackAdapt reported it, is close to unanimous. 91% say their organisation uses AI tools in marketing or advertising. 86% use them regularly or for most tasks. 88% report some AI-driven performance improvement. The regional spread is almost flat: 91% in North America, 90% in EMEA and 92% in APAC. There is no laggard region in this dataset that could be used to explain the rest of the findings away, which matters, because the interesting numbers in the study are the ones that stay low everywhere.
Comfort with AI holds until the human leaves the loop
StackAdapt asked respondents how comfortable they were with four different levels of AI involvement, and the answers form a ladder with one very steep step in it. The table below lists the four levels and the share of respondents comfortable with each, as the study reported them on 19 August 2026.
| Level of AI involvement | Marketers comfortable with it |
|---|---|
| AI recommends an action | 90% |
| AI prepares an action for human approval | 89% |
| AI acts within rules a human has set | 78% |
| AI operates autonomously once performance is proven | 50% |
The first two rungs are effectively the same answer at 90% and 89%. In both, the machine proposes and a person disposes. Letting AI act inside rules a human wrote costs 11 points and lands at 78%. Taking the person out of the loop, even with performance already proven, costs another 28 points and lands at 50%. Half of the profession will not delegate under the most favourable condition the question could have described. That is the shape of the gap: adoption is not the constraint, and neither is belief in the output, since 88% report performance improvement. The constraint is who gets to press the button.
Only 6% act on in-platform AI recommendations almost always
The single most quoted figure from the StackAdapt study is that 6% of marketers say they act on in-platform AI recommendations almost always. Inverted, that is 94% who do not act on them consistently. Google, Meta and Microsoft all ship recommendation surfaces inside their ad products, several with auto-apply switches that will implement the suggestion on the advertiser's behalf unless someone opts out. The study measures the demand side of that supply, and finds it thin.
Labelled analysis, not a study finding: the 6% is not evidence that platform recommendations are worthless. It is evidence that the recommendation surface, as currently built, does not clear the bar an experienced buyer applies before touching a live account. Those are different claims, and the study measures only the second one.
Why marketers ignore the suggestions: 42% say generic or irrelevant
StackAdapt asked the marketers who ignore in-platform recommendations why they ignore them. The reasons, as reported in the 19 August 2026 study, are below.
| Reason for ignoring the recommendation | Share of respondents |
|---|---|
| Feels generic or irrelevant | 42% |
| Does not align with strategy | 22% |
| Lack of explanation or transparency | 17% |
| Feels timed to drive spend rather than performance | 12% |
| Too many to evaluate | 6% |
| Other | 1% |
Two of those rows deserve separating out. The 12% who read recommendations as timed to drive spend rather than performance are making an accusation about incentive, and no amount of interface polish answers it. The 17% citing lack of explanation are making a much more tractable complaint: the suggestion arrived without its reasoning attached.
Labelled analysis: the 42% top answer, generic or irrelevant, is usually discussed as a trust problem. It reads more like a data-coverage problem. A recommendation engine writes its suggestion from what the platform can see, which is the account's own history, the platform's own inventory and modelled patterns from comparable advertisers. It cannot see the margin on the product, the stock position, the sales team's capacity, the pending price change or the fact that half the recorded conversions are junk. A suggestion built from a partial view of the business will read as generic to the person holding the whole view, and it will keep reading that way no matter how confidently it is phrased.
What does make marketers act: explanation, and a tie to a KPI
StackAdapt also asked what would make marketers act on a recommendation. 33% named a clear explanation or rationale. 31% named a clear tie to a KPI they care about. No other answer was reported at that level in the coverage of the release.
Those two answers describe the same missing artefact from two directions. A recommendation that says why it exists and which number it is supposed to move is a testable proposition. A recommendation that says only what to do is a request for trust, and the study shows that trust in the abstract is not the currency marketers are paying in. They already believe AI improves performance at 88%. They still will not act at 94%.
No dimension of the recommendation experience scored above 40% positive
StackAdapt rated the recommendation experience across five dimensions: relevance, timing, clarity, manageability and ease of action. Not one of the five scored above 40% positive. That is the quiet finding in the study, and it is arguably worse for the platforms than the 6% headline, because it says the failure is not concentrated in one fixable place. A relevance problem alone could be solved with better modelling. A clarity problem alone could be solved with better copy. Five mediocre scores across five independent dimensions describe a surface that was designed for a user who does not exist: someone with the authority to change spend, no obligation to justify the change, and time to work through a queue.
A checklist for reviewing an auto-applied recommendation before you accept it
The study does not prescribe a workflow. This checklist is our own, built from the two things the survey says make marketers act, which are a stated rationale and a link to a KPI.
- Read what the recommendation actually changes, not the label on it. Applying a "raise your budget" card changes pacing across every ad group underneath, including the ones you paused spend against on purpose.
- Find the KPI. If you cannot name the single number the change is supposed to move, and the direction, you are not evaluating a recommendation, you are accepting one.
- Check the conversion signal the suggestion is reasoning from. A recommendation built on a conversion action that fires on a thank-you page view, a duplicate tag or an imported goal nobody has audited will be internally consistent and externally wrong.
- Check what the platform cannot see. Margin, stock, lead quality downstream of the form, seasonality specific to your market, and any campaign that exists for a reason other than last-click return.
- Check the timing. A suggestion that arrives at the start of a promotion, mid-learning-phase, or three days after a tracking change is answering a question about a different account than the one you have now.
- Check whether auto-apply is on. In several ad products the default is that certain recommendation categories apply themselves. Review the change history rather than the recommendations tab, because the change history is where applied suggestions actually show up.
- Record the decision. If you reject a recommendation, note why, because the same card will return and the reasoning is worth more the second time.
Most of that work is measurement work rather than media work, which is why an account with clean, audited conversion tracking gets more value out of platform automation than an account without it. If the signal feeding the model is wrong, the recommendation is wrong with total confidence. A properly configured analytics and conversion setup is the precondition for delegating anything, and it is also the cheapest way to make platform suggestions less generic, because it widens what the platform can see. The same logic applies inside a Google Ads account, where auto-applied recommendations sit closest to live budget.
What the study did not say
This is a vendor-commissioned study. StackAdapt sells media buying technology, which means it has a commercial interest in a finding that says advertisers distrust the recommendations built into rival platforms. That does not make the numbers wrong. It does mean the framing was chosen by an interested party, and the study should be read as one input rather than as a settled account of the market.
The sample is 500 marketing and advertising professionals surveyed by NewtonX between 19 May and 8 June 2026, plus 187 StackAdapt clients surveyed separately. Five hundred is a reasonable professional sample and not a census. The client sub-sample is, by definition, people who already bought from the sponsor.
On geography, the study reports a three-region split only: North America, EMEA and APAC. It gives no Thailand breakdown and no Southeast Asia breakdown. Anyone quoting the 6% figure as a Thai statistic is inventing one. The study also does not report which platforms the recommendations came from, does not say how many of the surveyed marketers had auto-apply enabled, and does not measure what happened to accounts that did follow the recommendations. There is no performance outcome in this data, only stated behaviour and stated comfort.
What this means for Thai marketers
The auto-apply defaults that ship in Google Ads, Meta and Microsoft Advertising arrive in Thailand the same way they arrive everywhere else, and a Bangkok account manager sees the same cards as a New York one. What is not the same is the volume of local signal sitting behind those cards. Thai-language search behaviour, Thai conversion paths that end in LINE rather than a web form, and category volumes far smaller than the US equivalents all mean a recommendation engine has less to reason from before it makes a suggestion about a Thai account.
Reasoning from the study rather than reporting it: that is precisely the condition the 42% "generic or irrelevant" answer describes. The study did not measure Thailand, so this is an inference and should be treated as one. But the mechanism is not exotic. Thinner data produces more generic suggestions, and more generic suggestions get ignored, which is the loop the study documents in markets it did measure.
The practical consequence for a Thai advertiser is that the review step matters more here, not less, and that the fastest way to improve the quality of the recommendations you receive is to improve the quality of the signal you send. That means conversion actions that reflect real revenue, offline conversions imported where the sale closes offline, and a clear-eyed view of which of your logged conversions are actually chat clicks.
FAQ
Does this study say I should turn off auto-apply recommendations?
No, the study makes no recommendation about auto-apply settings at all. It reports that 6% of marketers act on in-platform AI recommendations almost always and documents the reasons the rest give for ignoring them. Whether auto-apply is right for a given account depends on the account's tracking quality and the reviewer's capacity, neither of which the study measured.
Is this data specific to Thailand?
No. The study reports a three-region split, at 91% adoption in North America, 90% in EMEA and 92% in APAC, and gives no Thailand or Southeast Asia breakdown. Applying the numbers to the Thai market is an inference, not a citation.
Who paid for the research?
StackAdapt commissioned it, and StackAdapt sells media buying technology. NewtonX fielded the survey of 500 marketing and advertising professionals between 19 May and 8 June 2026, and a parallel survey covered 187 StackAdapt clients. The commercial interest is worth holding in mind when reading the framing.
If 88% say AI improves performance, why does only 6% act on its recommendations?
Because the study measures two different things, belief in AI generally and willingness to act on a specific in-platform suggestion. The reasons given for ignoring suggestions are led by 42% saying the suggestion feels generic or irrelevant, followed by 22% saying it does not align with strategy and 17% citing a lack of explanation.
Do I have to do anything because of this study?
Nothing is required, since this is survey research and not a platform change. It is a reason to check whether auto-apply is enabled in your accounts and whether the conversion signal feeding those recommendations is trustworthy, which is worth doing regardless of what any survey reports.
The gap is a review problem, not an AI problem
Read together, the StackAdapt numbers describe a profession that has accepted AI as a tool and refused it as a decision-maker, and has fairly specific reasons for the refusal. 90% will take a recommendation. 50% will take an autonomous system. 6% will act on what the platform suggests almost always. The gap between those numbers is filled with review work, and review work is exactly what gets skipped when an account is under-staffed or the tracking is a mess. If you want help auditing what your ad platforms are recommending, and whether the data behind those recommendations is worth acting on, our team is happy to take a look.
