When to Trust the Google Ads Algorithm (And When Not To) - UK Guide 2026

The short answer

Trust the Google Ads algorithm when you have 30+ clean conversions per campaign per month, accurate conversion values, and a stable account. Don't trust it when you have low volume, dirty conversion data (junk leads, duplicate counting), or when you're testing something new. Smart Bidding is a force multiplier - it amplifies whatever signal you give it, good or bad.

The most expensive mistake Ben sees in UK audits is enabling tROAS or Maximise Conversion Value on a campaign with 12 conversions a month. The algorithm has nothing to learn from. It guesses, the guesses are bad, spend ramps up, and three weeks later the client wants answers.

How the Google Ads algorithm actually works

"The algorithm" is shorthand for several Google machine learning systems working together: Smart Bidding (tCPA, tROAS, Maximise Conversions, Maximise Conversion Value), broad match, asset rotation, audience expansion, and Performance Max's auto-allocation across channels. They share one dependency: conversion data.

Every conversion you report to Google Ads becomes training data. Google looks at the user's signals at the moment of the click - device, location, time of day, query, browsing history, audience signals, in-market segments, the works - and learns which combinations are most likely to convert. Then it bids more on those combinations and less on others.

This works brilliantly when:

• You feed it enough conversions to find patterns (statistical significance)
• The conversions you feed it are real (not junk leads or test fires)
• The conversion value reflects business value (not all leads worth £1)

It fails when any of those break. See Google's official Smart Bidding documentation for the underlying mechanics.

Conversion volume thresholds

Google's official guidance: 30 conversions per campaign in the last 30 days for tCPA, 50 conversions in the last 30 days for tROAS. Ben's real-world thresholds are higher because Google's minimums work on paper but produce noisy results in practice.

Critical: these are per campaign, not account-wide. An account with 200 conversions/month split across 10 campaigns has 20 conversions/campaign - too low for Smart Bidding on most of them. Either consolidate campaigns to hit the threshold, or run manual CPC on the small ones.

When to trust the algorithm

1. You have 50+ conversions per campaign per month

Above this threshold Smart Bidding consistently beats manual CPC for ~85% of UK accounts in Ben's experience. The signal is strong enough to find patterns humans can't.

2. Your conversion tracking is clean

One conversion = one real business outcome. No duplicate firing. No counting form-page-loads as form-submits. No counting page-views as conversions. Google Tag Assistant or conversion debugging tools can verify this in 10 minutes.

3. You're optimising for revenue, not leads

Ecommerce accounts with proper conversion value tracking are where Smart Bidding shines. tROAS on a Shopify store with 100+ orders/month routinely beats any human bidder. Google can see "this user buys £400 hairdryers, that user buys £20 brushes" and bid accordingly.

4. Your account is stable

No major budget changes, no new landing pages, no new product launches in the last 14 days. Smart Bidding needs a consistent learning environment. Constant change resets the learning period and produces volatile results.

5. You're using value-based bidding correctly

If different conversions are worth different amounts (a £5,000 enterprise lead vs a £50 SMB lead), tell Google. Pass dynamic conversion values, not flat values. See Google's value-based bidding setup guide.

When NOT to trust the algorithm

1. Low conversion volume

Under 30 conversions/month, Smart Bidding is guessing. The algorithm needs statistical patterns to learn. With 12 conversions, two random outliers (one massive sale, one fraudulent click) skew the model for weeks.

2. Junk lead problems

Lead-gen accounts where the form fills don't all become real opportunities. Google sees "100 conversions" and optimises towards them, but only 20 are real. The algorithm faithfully drives more of the same low-quality leads. Fix: import only qualified leads via offline conversion import (covered in Ben's separate post on offline conversion tracking).

3. New campaigns or major restructures

Brand-new campaigns have zero learning data. Starting them on tROAS is asking the algorithm to perform without a single training example. Start manual CPC for 30-60 days, accumulate data, then transition to Smart Bidding.

4. You can't measure incrementality

Smart Bidding will happily spend more on users who would have converted anyway (brand searches, retargeting). Without geo hold-out tests or proper incrementality measurement, you'll think Smart Bidding is winning when it's just claiming credit for organic demand.

5. Volatile or seasonal demand

If your business has wildly different demand week-to-week (event-based, weather-driven, pre-launch), Smart Bidding lags reality. By the time it learns the pattern, the season is over. Use seasonality adjustments or revert to manual CPC for spike periods.

6. Tight budget caps

If your daily budget is below 10x your average CPC, Smart Bidding throttles the campaign so much that learning never completes. Either raise the budget or use Maximise Clicks until the budget allows real Smart Bidding behaviour.

Smart Bidding strategies ranked

From "trust most" to "trust least" given proper conversion volume:

1. tROAS (with accurate conversion values) - the highest-leverage strategy when ecommerce data is clean
2. Maximise Conversion Value (with target ROAS) - same as above but Google sets the target
3. tCPA - reliable for lead gen with clean conversion tracking
4. Maximise Conversions (with target CPA) - good transition strategy from manual
5. Maximise Conversions (no target) - useful for new campaigns building data; risky long-term as Google chases volume regardless of cost
6. Maximise Clicks - traffic strategy, not a conversion strategy. Use for awareness or initial data collection only.

Performance Max: trust or override?

Performance Max forces you to trust the algorithm - that's the entire pitch. There is no per-channel bid adjustment, no manual CPC, no campaign-level placement control. You hand Google your conversions, your assets, and your audience signals, and it allocates spend across Search, Display, YouTube, Discover, Gmail and Maps.

When PMax works:

• Ecommerce with 50+ orders/month and good product feed
• Lead gen with offline conversion import sending qualified-lead values
• Strong audience signals (customer lists, custom segments matching your ICP)

When PMax doesn't:

• Low conversion volume (same threshold rules apply)
• Brand cannibalisation (PMax will eat branded traffic if not excluded - add brand keywords as account-level negatives via Google rep)
• You need transparency for client reporting (PMax channel reporting is still limited)

For the full PMax playbook see Google's official PMax guide.

Broad match with Smart Bidding

Google's pitch since 2022: "broad match + Smart Bidding finds queries you'd never think of". The pitch is half-true. Broad match works when paired with strong conversion data and a tight tCPA target. Without those guard rails, broad match is a budget vacuum.

Trust broad match if:

• Smart Bidding has 50+ conversions/month to constrain it
• You actively monitor search terms weekly and add negatives
• You're not in a category with high near-match ambiguity (e.g. "Apple" the company vs the fruit)

Don't trust broad match if any of those don't hold. Stick to phrase and exact match.

The hybrid approach Ben uses

For most UK accounts Ben manages, the structure looks like this:

• Brand campaigns: Manual CPC or Maximise Clicks with a low tCPA. Why? Brand traffic converts regardless - no need to pay Smart Bidding to "optimise" what's already free.
• High-volume non-brand: tROAS or tCPA, with accurate conversion values flowing in.
• Low-volume tests: Manual CPC for 30-60 days, then transition to Smart Bidding once volume threshold is hit.
• PMax: One PMax campaign per business goal, with branded terms negated, audience signals supplied, and value-based bidding on.
• New product/landing page tests: Manual CPC during the test, then move to Smart Bidding once a winner is chosen.

The principle: trust the algorithm where it has data and the data is clean. Override it where it doesn't. Never blindly switch every campaign to Smart Bidding because Google's recommendation engine suggested it (see Ben's separate post on why most Google Ads recommendations should be ignored).