Ghost Ads

What Are Ghost Ads?

Ghost ads are suppressed bids entered into ad auctions on behalf of users assigned to a holdout control group — bids that participate in the auction environment but are deliberately designed to lose, ensuring the user never sees the advertisement.

Also referred to as ghost bidding or bid suppression, this technique is the measurement infrastructure behind platform-native incrementality studies such as Meta’s Conversion Lift.

The purpose is precise: construct a clean control group for holdout testing without removing those users from the bidding pool.

In a standard holdout design, the control group is simply excluded from targeting. Ghost ads take a more sophisticated approach — holdout users remain participants in every auction, the platform just ensures it loses those auctions every time. The holdout group stays unexposed. The auction stays intact.

How the Mechanism Works

Understanding ghost ads requires understanding what they are designed to prevent: auction deflation.

The Auction Deflation Problem

In any real-time bidding environment, CPMs are set by competitive pressure among bidders. When a standard holdout removes 20% of the target audience from the auction, competition drops. CPMs for the remaining 80% of exposed users fall artificially — sometimes by 10–25% depending on audience density.

This creates a measurement contamination: the exposed group receives ads at lower CPMs than they would in a live campaign at full scale. The comparison between exposed and holdout groups is no longer apples-to-apples. Your incremental lift estimate is inflated by a cost efficiency advantage that wouldn’t exist outside the experiment.

How Ghost Bidding Solves It

Ghost bidding keeps all users — including holdout users — inside the auction. For control group users, the system enters a bid at $0 or the minimum technically permissible floor. It participates, loses, and logs the result.

Because holdout users are still competing in auctions, the supply-demand dynamics for exposed users remain unchanged. CPMs stay at true market rates. The experiment now reflects what would actually happen at full campaign scale — making the incremental lift estimate valid for budget projection.

What the Control Group Experiences

Holdout users see no ad from the campaign being tested. They may see organic content, ads from other campaigns, or competitor ads — none of which are suppressed. The ghost bid only controls for the specific campaign under measurement.

This is important for lead attribution: a holdout user who converts does so through channels other than the tested campaign, providing the organic baseline conversion rate against which the campaign’s incremental lift is calculated.

Ghost Ads vs. Standard Holdout Testing

Both methods attempt to isolate incremental performance. The structural difference in how they construct the control group produces meaningfully different measurement outcomes.

Dimension	Ghost Ads	Standard Holdout
Auction participation	All users bid; holdout bids lose	Holdout users excluded from bidding
CPM accuracy	True market rate preserved	Artificially deflated for exposed group
Incremental lift accuracy	High — auction-neutral control	Moderate — CPM bias distorts estimate
Platform dependency	Requires native platform support	Executable on any platform
Implementation complexity	Higher — platform-managed	Lower — audience exclusion list
Holdout contamination risk	Low	Moderate (organic impression bleed)
Primary use case	Paid social, programmatic incrementality	Cross-channel, geo-based tests

For high-volume paid social lead generation programs where CPM variance meaningfully impacts CPL, ghost ads are the structurally superior holdout design. For cross-channel or geo-based incrementality tests, standard holdout methods remain the practical choice since ghost bidding infrastructure is not universally available.

Measuring Incremental Lead Generation

Ghost ad experiments produce the same core metrics as any holdout test — but with higher confidence in the causal validity of the estimate.

The Core Lift Formula

iLift (%) = [(CVR_exposed − CVR_ghost_holdout) / CVR_ghost_holdout] × 100

Where:
  CVR_exposed       = conversion rate of users who saw the campaign
  CVR_ghost_holdout = conversion rate of users in the ghost ad holdout cell

Incremental Leads = (CVR_exposed − CVR_ghost_holdout) × Total Exposed Users

Because the ghost holdout preserves auction dynamics, CVR_ghost_holdout reflects a true organic baseline — not one inflated by favorable CPM conditions that wouldn’t exist at scale.

Connecting Ghost Ad Results to Lead Attribution

The value of ghost ad experiments compounds when holdout membership is tracked at the lead level inside your CRM.

When a lead from the holdout group submits a form, their lead source data — channel, campaign, UTM parameters — captures what actually drove their conversion. Comparing the lead source distribution of holdout converters against exposed converters reveals which organic channels are generating baseline demand and which are being incorrectly credited in your attribution model.

This lead-level view transforms ghost ad data from a channel-level incrementality estimate into a granular attribution correction tool — identifying exactly where your multi-touch model is overcounting paid influence and where organic is doing more work than your dashboards show.

Limitations and Considerations

Ghost ads solve the auction deflation problem but introduce their own operational constraints.

• Platform availability is limited. Native ghost bidding is supported by Meta (Conversion Lift studies), some programmatic DSPs, and select demand-side platforms. Google Ads does not natively support ghost bidding; Google Campaign Experiments use a different holdout architecture. Teams running cross-platform tests cannot apply ghost bidding uniformly across channels.
• Organic exposure is not controlled. Ghost bidding suppresses only the tested paid campaign. Holdout users still see organic search results, social content, PR coverage, and competitor ads. In high-brand-visibility categories, this organic exposure can meaningfully close the gap between holdout and exposed conversion rates, compressing the measured incremental lift.
• Minimum audience thresholds apply. Ghost ad holdout groups require sufficient size to reach statistical significance. Meta’s Conversion Lift requires a minimum audience of approximately 10,000 users per cell before the experiment will generate reportable results. For niche B2B audiences with tight ICPs, this threshold may be difficult to reach within a single campaign.
• Budget allocation to ghost bids. While ghost bids are designed to lose, entering auctions consumes bidding infrastructure capacity. On some platforms, ghost bid activity marginally reduces the impression budget available to the exposed group. In tightly paced campaigns, this can introduce a small downward bias in exposed group reach.
• Results are campaign-specific, not channel-level. A single ghost ad experiment measures the incrementality of one campaign — not the entire channel. Channel-level incrementality requires either a full channel holdout (no campaigns running to holdout users) or aggregating results across multiple ghost ad experiments over time.

Best Practices for Holdout Validity

Ghost ad experiments produce defensible data only when the experimental design is rigorous from the start.

• Size the holdout cell at 10–20% of the target audience. Below 10%, the holdout group rarely accumulates sufficient conversion volume to reach statistical significance within a 2–4 week test window. Above 25%, you sacrifice meaningful reach and CPL performance during the test period without proportional measurement benefit.
• Lock holdout assignment before the campaign launches. Assigning users to the holdout cell after observing early results introduces selection bias. The assignment mechanism must be randomized and sealed before any impressions are served.
• Run experiments for a minimum of two full business cycles. A single week of ghost ad data captures day-of-week conversion variance rather than genuine behavioral signal. Two to four weeks is the industry-standard minimum; B2B programs with 30–60 day consideration cycles should run ghost ad windows of 4–8 weeks to capture downstream MQL and SQL conversions.
• Validate parallel trends in the pre-period. Before declaring results valid, confirm that holdout and exposed groups had equivalent conversion rates in the 1–2 weeks before the experiment launched. Non-parallel pre-period trends indicate audience imbalance that will contaminate the lift estimate regardless of ghost bid quality.
• Integrate holdout group membership with your lead attribution platform. Pass holdout cell membership as a custom parameter to your CRM at the point of form submission. This enables full-funnel analysis: not just whether holdout leads converted at a lower rate, but whether they progressed through the pipeline — MQL to SQL to opportunity — at a different velocity. Top-of-funnel lift alone is insufficient for B2B budget decisions.
• Do not run concurrent campaigns to holdout users from other campaigns. If holdout users are simultaneously exposed to other live campaigns for the same brand, their organic baseline is contaminated. True ghost ad isolation requires that holdout users receive no paid brand exposure during the test window — not just from the tested campaign.

Frequently Asked Questions

Are ghost ads the same as holdout testing?

Ghost ads are a specific holdout testing technique — not synonymous with holdout testing broadly. Holdout testing is the general methodology of withholding a campaign from a control group and comparing outcomes. Ghost ads are the mechanism used to implement that holdout while preserving auction dynamics. Standard holdout testing excludes control users from the bidding pool entirely; ghost ads keep them in the pool but ensure they never win an impression.

Which ad platforms support ghost bidding natively?

Meta (Facebook and Instagram) is the most widely used platform with native ghost bidding infrastructure through its Conversion Lift product. Some programmatic DSPs — including The Trade Desk and certain DV360 configurations — support ghost bid holdout designs. Google Ads uses a different holdout architecture in its Campaign Experiments, which randomizes at the campaign level rather than the user level. Most search and email platforms do not support ghost bidding and require geo-based or time-based holdout designs instead.

How do ghost ads affect CPL during the experiment?

Ghost ad experiments may marginally increase CPL during the test window for two reasons: the holdout cell reduces effective audience reach (10–20% of users are withheld from conversion), and ghost bid activity consumes a small share of bidding capacity. These effects are typically small — 3–8% CPL elevation — and are the acceptable cost of generating a statistically valid incrementality estimate. Budget for this variance in test-period performance expectations before presenting results to finance.

Can ghost ad experiments measure B2B lead quality, not just volume?

Yes — and this is the highest-value application for B2B revenue marketers. By passing holdout cell membership into your CRM alongside standard lead source data, you can compare MQL-to-SQL conversion rates, average deal size, and sales cycle length between exposed and holdout cohorts. A channel that generates high MQL incrementality but low SQL incrementality is producing low-quality incremental demand — a finding that changes the budget case for that channel entirely.

What is the difference between a ghost ad experiment and a dark period test?

A dark period test withholds a campaign from the entire audience for a defined period, then compares performance against a period when the campaign ran. Ghost ad experiments run simultaneously across exposed and holdout groups within the same time window. Dark period tests are simpler to implement but introduce temporal confounds — seasonal demand shifts, competitor activity, or algorithmic learning resets can distort the comparison. Ghost ad holdout designs eliminate temporal confounds by running both groups concurrently.

How many users need to be in the holdout group for valid results?

The required holdout size depends on your baseline conversion rate and the minimum detectable effect (MDE). At a 1.5% lead form conversion rate with a target MDE of 20% relative lift, each cell requires approximately 6,500 users at 95% confidence and 80% power. For Meta Conversion Lift specifically, the platform enforces a minimum of roughly 10,000 users per cell before generating reportable results. Size the holdout percentage against your total addressable audience to hit these thresholds within your test window — do not reduce the holdout percentage to preserve reach at the cost of statistical validity.

Do ghost ads work for retargeting campaigns?

Ghost ad holdouts for retargeting require additional design care. Retargeting audiences are inherently high-intent — users who have already visited your site or engaged with prior content. Because these users have elevated organic conversion rates, the holdout baseline will be higher than for prospecting campaigns, compressing the measured incremental lift. This does not mean ghost ads are ineffective for retargeting measurement — it means the iLift threshold for declaring retargeting incremental must account for the high organic baseline. Use the ghost holdout CVR from the retargeting audience specifically, not your overall site conversion rate, as the baseline comparator.