Monetizing GenAI Without Tanking Search Engagement: The Playbook

Monetizing GenAI Without Tanking Search Engagement: The Playbook

Introduction

If you’ve used ChatGPT (or any GenAI app) lately, you’ve probably noticed the quiet problem underneath the magic: running these systems costs a lot, especially for search-style interactions. So GenAI firms are hunting for monetization strategies beyond “just subscriptions,” and the big question is: how do you recover costs without killing user engagement?

New research from Veronica Rosendo-Rios and Paurav Shukla (based on the original paper) digs into exactly that. The study looks at how advertising-supported monetization can affect users’ intent to upgrade/downgrade—and how that choice connects to search query engagement once ads enter the experience.

The researchers build their framework using two psychology ideas: the compromise effect (people often like middle options when forced to choose) and affective primacy (what you feel emotionally can shape what you think and what you do next). Across four experiments (total N = 1063), they test what happens when GenAI firms add one or even two ad-supported tiers—and when ad “intrusiveness” (timing/length vs. visual style) changes how users engage with search queries.

Why This Matters

This is significant right now because GenAI monetization isn’t just a pricing problem—it’s an experience design problem. Most users still start on free tiers, while the costs of serving GenAI queries can be enormous (the paper cites estimates that GenAI queries can be up to 10× more expensive than traditional search). When a company tries to fix revenue with subscriptions alone, the conversion rates are typically low—meaning the business is stuck with a huge “free user base” and mounting compute costs.

What makes this research practical is that it treats monetization as something users experience moment-to-moment. Instead of only asking, “Will people pay?”, it asks, “Will people still engage with the search task after ads show up?” That’s a very real threat: if ads reduce engagement, you don’t just lose ad revenue—you can also weaken retention, power, and future willingness to pay.

A scenario where this can apply today: imagine a GenAI search assistant that plans to roll out an ad-supported tier. They’ll be tempted to choose the most aggressive ad formats (“bigger visuals!” “shorter interruptions!”). But this research suggests the opposite might be true—especially for temporal intrusion (ad length), where the paper reports a counterintuitive finding: longer ads led to higher engagement in their experimental setup. That’s the kind of insight that can save a monetization rollout from accidentally harming the core “search usefulness” feeling.

This also builds on and extends earlier AI platform research. Prior work often focuses on how GenAI improves marketing automation or changes platform value—but this paper connects monetization architecture directly to user decision psychology and task engagement, using a measurable outcome: user engagement with a specific search query inside a simulated GenAI interface (not just general attitudes).

How Compromise Pricing Shapes Upgrade vs. Downgrade Behavior

The paper’s core strategic lens is the compromise effect: when users face extremes (free with limited features vs. paid with no ads), they tend to like a middle option. In this context, the “middle” is an advertising-supported tier that offers fuller capabilities without full subscription pricing.

Crucially, the researchers test not only whether users like the ad-supported option, but also whether it behaves differently depending on whether someone is already a paid subscriber. That matters because a firm isn’t just trying to monetize free users—it’s also trying not to trigger mass downgrades from current subscribers.

What the researchers compared (single compromise option vs. subscription-only)

In Study 1, they compared:
- a single advertising-supported option (the compromise)
- against a baseline where users choose between free vs subscription only

Here’s the option structure used in Study 1:

Sample size: Study 1 used 306 participants from the US (after excluding one who failed checks), with a mean age around 38.17. Paid subscribers were 16.1% of the sample.

The headline result: ads win as the middle option

When the ad-supported tier was introduced, preference shifted hard toward the compromise. In the experimental condition:
- Ads tier: 70.22% preferred
- Free tier: 18.22%
- Subscription: 11.56%

In the control condition (no ads tier):
- Free: 66.3%
- Subscription: 33.7%

Even more interesting: the compromise effect appeared for both segments.
- Free users strongly favored the ad tier (72.9%)
- Paid subscribers also picked the ad tier more than expected (54.5%), meaning ads can create downgrade risk

That’s the core trade-off the paper keeps coming back to: ads can monetize free users—but may cannibalize subscription revenue if the “choice architecture” isn’t handled carefully.

Why Adding a Second Ad-Supported Option Changes Everything

If Study 1 taught us “ads can pull people into the middle,” Study 2 teaches us the next lesson: real pricing pages rarely show only one compromise option. When platforms add multiple tiers, decision difficulty rises, and that changes behavior differently for free users versus paid users.

This connects directly to the compromise effect literature: more choice options can weaken the simple middle-option appeal. But in GenAI monetization, that weakening may not be bad—it can be segment-specific.

Study 2 tested what happens with multiple compromise models

Study 2 added another ad-supported tier that combined:
- ads at a different interval
- plus a low subscription fee in one option

The paper describes four options in Study 2, including:
1. Free (no ads) with limited access/features
2. Ads (regular intervals, full capabilities, latest engine)
3. Ads + subscription ($10/month) with ads and full capabilities
4. Subscription without ads ($20/month)

Here’s the comparison framing the paper used (simplified to match the described incentives):

Sample size: 309 recruited, 302 after attention checks.

Results: the compromise effect becomes asymmetric across segments

Compared to the control condition:
- Control: most people preferred free (68.3%) then subscription (28.7%)
- Experimental: the top preference was ads at regular intervals (30.3%), then free (26.6%)

But the segmentation is what matters:
- For free users, the ad-supported option is still the best “middle,” and they’re more likely to upgrade.
- For paid subscribers, offering an additional ad-supported compromise makes them less likely to downgrade relative to the single compromise setup.

So the paper’s central “architecture” insight becomes clear:
multiple compromise options increased free users’ upgrade intentions and reduced paid users’ downgrade risk.

Practical implication

If you’re a GenAI firm trying to add ads, don’t just think “one ads tier vs no ads.” The research suggests you should think in terms of tier menus:
- a single ad tier may unintentionally make downgrades attractive for existing subscribers
- a more nuanced set of options can steer free users toward ads without making paid feel like a worse deal

Visual vs. Temporal Intrusion: Why “Bigger Ads” Aren’t the Main Problem

After establishing that advertising tiers can shift upgrade/downgrade behavior, the paper asks a more operational question:
once an ads tier exists, does it reduce engagement with the user’s search task?

They test ad intrusiveness in two dimensions:
- visual intrusiveness (text vs image vs video format)
- temporal intrusiveness (ad duration/length)

This matters because user engagement is what protects the “search usefulness” loop. If ads disrupt the task, the platform risks losing the thing that makes users come back.

Study 3: visual format didn’t significantly change engagement

In Study 3, participants (UK-based; N = 208 after exclusions) imagined a ChatGPT-like interface and performed a search query. Then they were shown one ad format:
- text ad (Google-style results)
- image ad (Instagram-like)
- video ad (YouTube-like, shown statically to control motion)

Key point: a pilot and the main study indicated no significant difference in perceived visual intrusiveness across formats.

So the moderation hypothesis around visual intrusion wasn’t supported:
- visual intrusion didn’t significantly affect engagement
- the predicted mediation path through affective → cognitive evaluations was not statistically significant in the model they tested (though they mention an approximately 34.8% contribution in an indirect accounting sense)

Why might this be happening?

The authors float an explanation that feels plausible: user desensitization. In heavily commercialized digital settings, users may have learned to filter visual ad noise. If so, format alone (text vs image vs video) might not be the lever that changes how users feel during the search task.

Practical take: don’t over-index on “make ads smaller” or “make them look like search results” as your only engagement strategy.

Temporal Intrusion Can Increase Engagement (Yes, Really)

Study 4 is where the research gets most counterintuitive—and most actionable for product teams.

Instead of visual format, Study 4 manipulates ad length:
- short: 6-second ads
- long: 30-second ads
Participants encountered ads that were unskippable to simulate realistic environments.

Sample size: 228 participants (UK; after excluding 12 failed checks).

Study design: where the ads appear relative to search

They used a 2×2 setup:
- control vs advertising-supported model placement
- and ad duration low vs high (6s vs 30s)

In the advertising-supported version:
- ads were shown immediately after the search query and before results

In the control condition:
- participants still saw ads, but the ads were shown before or after the searches (i.e., not positioned as an immediate interruption to the query→results loop)

The big moderation finding

When looking at engagement with the search task, ad length interacted with the “ads interruption” condition in a surprising way:

• In the control condition, users preferred shorter ads over longer ones:

  • short ads mean engagement score: 2.85
  • long ads mean engagement score: 1.94

• In the advertising-supported condition (ads interrupting the query→results flow), users preferred longer ads:

  • short ads mean: 2.16
  • long ads mean: 2.46

So temporal intrusion didn’t behave like “longer = worse.” Instead, the longer ads were more tolerable (and associated with higher engagement) when they were experienced as part of the ad-supported tier.

How affective primacy explains it

The paper supports an affective primacy story using a serial mediation model:
1. temporal intrusion (via ad-supported placement + length) affects affective evaluation (emotions)
2. affect then shapes cognitive evaluation
3. cognitive evaluation predicts search query engagement

They also explicitly test the reverse order and find it unsupported—supporting affective primacy rather than cognition-first.

In their reported moderated mediation analysis:
- the interaction between ad-supported model and temporal intrusion significantly predicted affective evaluation
- affective evaluation significantly predicted cognitive evaluation
- cognitive evaluation predicted search query engagement

Practical implication you can use immediately

If you’re designing ads for GenAI, don’t assume shorter ads are always better for engagement—especially if the ad disrupts the immediate interaction loop.

Instead, the paper suggests a design principle:
- brief interruption may feel abrupt and “empty,” triggering immediate negative affect
- longer interruption may provide enough content/context to reduce frustration and support better emotional processing

That’s not a blanket rule for every platform, but it’s a strong signal that ad timing and “how ads fit the conversation” can outweigh ad format styling.

Key Takeaways

Key Takeaways

• A single advertising-supported tier acts as a “compromise” and can strongly increase preference—especially for free users. In Study 1, the ads tier was preferred by 70.22% of participants in the experimental condition.
• But a single ad tier can create downgrade risk: even paid subscribers leaned toward the ads option in Study 1 (54.5% preferred ads over subscription).
• Offering multiple ad-supported compromises changes the outcome: Study 2 suggests that adding a second compromise option encourages free users to upgrade while reducing downgrade intentions among paid subscribers.
• Visual ad style (text vs image vs video) didn’t significantly impact engagement in Study 3, and the predicted visual-intrusion mechanism wasn’t supported.
• Temporal intrusion (ad length) did matter—and in a counterintuitive way. In Study 4, when ads interrupted the query→results flow, longer ads (30s) were associated with higher search engagement than shorter ads (6s).
• The engagement mechanism is consistent with affective primacy: ad intrusiveness shifts users’ initial emotions, which then shape cognitive evaluation, driving engagement.
• For product teams: don’t treat ad monetization as only a “revenue add-on.” Treat it like choice architecture + emotional UX design, especially around when ads appear in the search journey.

If you want, tell me what kind of GenAI app you’re thinking about (chat assistant, search summarizer, coding tool, etc.), and I can translate these findings into a concrete tiering + ad-placement experiment plan.