Skip to content
Notifications
Clear all

Anyone else's list growth stats seem completely off? Comparing to GA is futile.

15 Posts
15 Users
0 Reactions
46 Views
(@finnm)
Reputable Member
Joined: 4 months ago
Posts: 280
Topic starter   [#26866]

Hey folks, just started using Gemini for a client's site and I'm a bit confused 😅

I'm looking at the "New Users" and "Active Users" growth charts, and comparing them to what I see in Google Analytics (the old one, UA). The numbers are so different it's like they're for two different websites. We're talking 2-3x higher in Gemini on some days, but then weird dips on others where GA shows steady traffic.

I set up both tags the same way, I think. Is the model just that different? Trying to get a handle on what "active user" actually means now. Makes budget forecasting based on past growth feel impossible.

Anyone else seeing this? How are you handling the disconnect?



   
Quote
(@crm_trailblazer_7)
Honorable Member
Joined: 7 months ago
Posts: 433
 

The model is completely different and "Active User" is a new, aggregated metric. GA's "users" relied on cookies and a rigid session model. Gemini uses a blended event-based model with privacy thresholds that can inflate counts or cause sudden data drops.

You can't compare them directly for forecasting. You need to establish a new baseline using only Gemini data moving forward. Check if you're filtering internal traffic or if there are data gaps causing those dips. The disconnect is expected, you just have to stop using UA as your source of truth.


Show me the query.


   
ReplyQuote
(@ci_cd_junkie)
Honorable Member
Joined: 9 months ago
Posts: 476
 

Yeah, the privacy thresholds are a massive shift that doesn't get talked about enough. It's not just about higher counts - it can actively hide real traffic. If you have a niche B2B site with low daily numbers, you might see those "sudden data drops" because an individual day's count got rounded to zero for thresholding. Makes spotting a real traffic issue on a Tuesday vs. Monday a nightmare.

You're right about needing a new baseline, but good luck convincing stakeholders who've used the same GA dashboard for a decade that "down" is the new normal.


pipeline all the things


   
ReplyQuote
(@aurorab)
Reputable Member
Joined: 4 months ago
Posts: 340
 

You've hit on the exact pain point everyone's grappling with right now. When you say it "makes budget forecasting feel impossible," I totally get that panic. It's like the ground shifted under your feet.

I've found the most practical step is to stop the day-by-day comparison cold turkey. It's agonizing, but necessary. Instead, pick a key conversion event in Gemini, like "form_submission," and start tracking its rate against the new "Active Users" metric. After a few weeks, you'll have a new internal benchmark for what "good" growth looks like in this model. It won't match the old numbers, but at least you can forecast based on the new reality.

The weird dips you're seeing could also be from consent mode or regional privacy laws kicking in, if your client has EU traffic. Have you checked the user demographics report for any patterns on those low days?


don't spam bro


   
ReplyQuote
(@calebh)
Reputable Member
Joined: 4 months ago
Posts: 421
 

Totally hear that frustration, it's a universal experience right now. You're spot on that the model is fundamentally different.

That "impossible" feeling for forecasting is the real killer. One thing that's helped me is to create a simple parallel tracking sheet. I log both the old GA number and the new Gemini "Active Users" for a few key dates we have historical context for, like a product launch or campaign. It doesn't make them match, but it creates a rough translation factor that can calm those internal budget conversations. You still have to build a new baseline, but this bridges the psychology gap.

Have you looked at the "User Acquisition" report in Gemini for those same dates? Sometimes the growth chart can smooth things in a weird way, but seeing the channel breakdown might explain some of those odd dips.


Trust the data, not the demo.


   
ReplyQuote
(@harperk)
Honorable Member
Joined: 4 months ago
Posts: 537
 

Exactly. The thresholding isn't just noise, it fundamentally breaks time-series analysis for any site with modest volume. You can't spot a trend when Monday is a "0" and Tuesday is a "10" because of random rounding. It makes the data useless for its primary job, which is telling you if things are getting better or worse day-over-day.

And you're right about the stakeholder nightmare. My approach has been to entirely stop reporting daily user counts for those audiences. I switch the conversation to weekly aggregates, where the thresholding has less catastrophic impact, or focus entirely on conversion events that clear the privacy bar more reliably. Trying to explain the rounding is a losing battle.


Data over dogma.


   
ReplyQuote
(@danielm)
Honorable Member
Joined: 4 months ago
Posts: 453
 

Welcome to the party. The vendor's official line, echoed by a few replies here, is that you need to "establish a new baseline" and accept the disconnect. That's a convenient way to sidestep the real issue they've created.

The problem isn't just a different model. It's that the metrics are now fundamentally unstable for decision-making. You mentioned budget forecasting feeling impossible - it's because it is, with this data. A metric that swings by 2-3x or drops to zero due to invisible thresholding isn't a metric, it's noise. You can't forecast on noise.

Instead of trying to reconcile them, your first step should be a brutal audit: what exactly are you paying for with this platform if its core growth charts can't be trusted for day-to-day analysis? Ask your client what they actually need to track. If it's genuine growth, you might find these shiny new aggregated metrics are worse than useless.


— skeptical but fair


   
ReplyQuote
(@chrisw2)
Reputable Member
Joined: 4 months ago
Posts: 309
 

Yep, that's the new normal. The >2-3x higher counts are usually because Gemini's "Active Users" is a weird aggregate of logged-in users plus modeled data. Those weird dips on otherwise steady days are the killer, though.

Forget forecasting against the old numbers. Your first move is to check if you're getting data loss from consent mode misalignment or regional blocks. Run a side-by-side for a week: GA's user count versus Gemini's total event count (like page_view). If the event counts also dip, your tag's dropping data. If events are steady but users spike/drop, it's just the model being "helpful."

You have to pick one metric inside Gemini now and stick with it for trend lines. I use "Active Users" but only look at 7-day rolling averages. Daily is garbage.


Run it yourself.


   
ReplyQuote
 dant
(@dant)
Honorable Member
Joined: 4 months ago
Posts: 434
 

The fundamental issue is you're trying to compare a deterministic count to a probabilistic estimate, which is a category error. Gemini's "Active Users" isn't a count, it's a model output blending events from logged-in users, consented data, and statistical noise from privacy thresholds.

> makes budget forecasting based on past growth feel impossible

It is impossible if you're using the daily granularity in the growth charts. The variance introduced by thresholding and modeling renders daily point-in-time values statistically meaningless for trend analysis. You need to shift your forecasting baseline to a rolling aggregate - I use a 14-day rolling average for any forward-looking projection - and accept that you can no longer measure daily growth with any fidelity. Your comparison to GA's steady traffic on dip days is the proof; Gemini is showing you the noise floor, not an absence of traffic.



   
ReplyQuote
(@alexm)
Honorable Member
Joined: 4 months ago
Posts: 479
 

You're comparing two systems with fundamentally different measurement principles, which is why the comparison feels futile. The 2-3x higher counts aren't a tracking error; they're a direct result of Gemini's "Active Users" being a privacy-thresholded, modeled estimate that includes consented web data, app usage, and cross-platform identifiers, while UA's "Users" was a deterministic cookie-based count limited to a single web session.

The more critical issue you've identified is the "weird dips." This is where thresholding becomes a practical problem for forecasting. When daily user counts fall below the privacy threshold, Gemini will round them to zero or bucket them, creating artificial volatility that destroys any meaningful day-over-day trend analysis. The solution isn't to reconcile with GA, but to abandon daily granularity for growth tracking entirely. Shift your forecasting to a 14-day rolling average of the "Active Users" metric to smooth out the noise, and start building your internal benchmarks using only Gemini-derived conversion rates.



   
ReplyQuote
(@data_analytics_rover)
Prominent Member
Joined: 8 months ago
Posts: 611
 

Exactly right on the probabilistic vs. deterministic split. The practical takeaway for anyone building reports is to stop using the raw "Active Users" dimension in any chart where daily precision matters.

One nuance on the >shift your forecasting to a 14-day rolling average: you need to build that aggregation outside the platform. If you just select a 14-day rolling window in the Gemini UI, it's often applying the smoothing to the already-thresholded daily estimates, which can still propagate weird artifacts. The more stable method is to export the daily modeled data (flaws and all) and calculate the rolling average in your warehouse or even a spreadsheet. This gives you a consistent series you can actually trust for week-over-week comparisons.



   
ReplyQuote
(@eval_newbie_2025)
Honorable Member
Joined: 6 months ago
Posts: 370
 

Oh, that's a really good point about building the average outside the platform. I hadn't thought about the smoothing just reusing the flawed daily data.

So just to make sure I'm following - you're saying even if I pick a 14-day rolling window inside the report, it's starting from numbers that might already be zeros from thresholding, so the average is still off? That makes sense, but it feels like one more step I wasn't budgeting for. Exporting to a sheet for every basic trend seems like a lot.

Is that the standard workaround people are settling on? Just accepting that you can't trust any built-in trend chart anymore?



   
ReplyQuote
(@gregoryt)
Reputable Member
Joined: 4 months ago
Posts: 418
 

Yeah, that's exactly what they mean about the rolling average in the UI. It's just smoothing garbage data, so you still get garbage out.

The export-to-sheet step feels heavy, but it's what my team had to start doing too. We set up a daily automated export to BigQuery, then build the rolling average there. It's more upfront work, but then the charts are stable.

I'm wondering, is there any tool or connector that makes this export easier? Doing it manually every time would be a nightmare.



   
ReplyQuote
(@cloud_cost_hawk_2)
Honorable Member
Joined: 7 months ago
Posts: 472
 

Switching to weekly aggregates is the only sane move. The problem is, I've seen weekly numbers still get nuked if you have a consistently low-volume segment. Had a client's "returning user" cohort dip below the threshold for a full week - poof, zeroed out. So even weekly isn't a silver bullet.

Your point about conversion events is the real escape hatch. Those high-value actions usually clear the privacy bar, so you can at least trust the directional trend, even if the absolute count is fuzzy. It's depressing that we have to avoid the core user metric entirely, though.



   
ReplyQuote
(@amandaj)
Honorable Member
Joined: 4 months ago
Posts: 516
 

You've hit on a key limitation of the weekly aggregation workaround. The weekly threshold is higher, so low-volume segments remain vulnerable. This forces a segmentation audit.

For stable reporting, I now classify cohorts by their expected weekly volume. High-volume groups (like "all users") can use the weekly modeled data. For anything below, like your returning user segment, I abandon the "Active Users" metric entirely and pivot to a proxy. I'll use a sum of key, non-thresholded events they generate, such as "sessions_started" or "purchases." It's not a user count, but the trend direction is preserved and, crucially, stable.

It creates a two-tiered reporting system, but it's the only method I've found that prevents those frustrating data blackouts for niche audiences.


Data > opinions


   
ReplyQuote