Why conversion tracking is the foundation of everything else
Smart Bidding strategies — Maximize Conversions, Target CPA, Target ROAS — all work by predicting which auctions are likely to lead to the conversion action you have defined and bidding accordingly. If that conversion action is being tracked incorrectly, the algorithm is optimizing toward the wrong target with full confidence, and it will not tell you something is wrong. It will simply produce results that look plausible but are built on bad data.
The most expensive conversion tracking mistakes are not the obvious ones (tag not firing at all is usually caught quickly because conversions drop to zero). The expensive ones are subtle: a tag firing twice on certain browsers, a "Thank You" page that is reachable without completing a purchase, or a lead form confirmation that fires regardless of whether the form was actually submitted successfully.
Setting up conversion tracking with Google Tag and GTM
Google Tag (gtag.js) is the base tag that connects your website to Google Ads, GA4, and other Google products. It can be installed directly on your site or — for most teams, preferably — managed through Google Tag Manager (GTM), which lets you add, edit, and test tags without touching site code for every change.
In GTM, a Google Ads Conversion Tracking tag fires the conversion event to Google Ads. The trigger determines when it fires — typically a page view on a confirmation/thank-you page, or a custom event fired by your site's JavaScript when a form submission or purchase completes successfully.
The single most common setup error is using a "page view" trigger on a URL that can be reached without completing the action — for example, a thank-you page accessible by direct URL, browser back button, or page refresh. Each of those reloads counts as another conversion, inflating your numbers and confusing Smart Bidding.
- Prefer event-based triggers (form submit success, purchase confirmation event) over page-view triggers where possible
- For page-view triggers, ensure the confirmation page cannot be reached without completing the action — check what happens on refresh and back-button
- For purchase conversions, pass transaction value and currency dynamically — never hardcode a fixed value for variable-priced purchases
- Use GTM's Preview mode to verify a tag fires exactly once per genuine conversion before publishing
Enhanced conversions and conversion modeling
Enhanced conversions improve measurement accuracy by sending hashed first-party customer data (email, phone, name, address) alongside the conversion event. This allows Google to match conversions to ad clicks even when third-party cookies are blocked or unavailable — which is now the default in most major browsers.
Enhanced conversions for web require sending this data via gtag.js or GTM with the customer data automatically hashed (SHA-256) before it leaves the browser — Google never receives unhashed personal data. Enhanced conversions for leads extends this to offline conversions, matching CRM-reported leads back to the original ad click using hashed contact data uploaded after the fact.
Conversion modeling fills measurement gaps statistically when direct tracking is not possible (consent declined, cross-device journeys). It is enabled automatically when enhanced conversions are active and consent mode is configured, and it generally improves reported conversion volume accuracy without requiring additional setup beyond the base implementation.
GA4 imported conversions vs native Google Ads conversions
You can track conversions in Google Ads two ways: a native Google Ads conversion action (tag fires directly to Google Ads), or a GA4 conversion event imported into Google Ads through the linked GA4 property. Both are valid, but mixing them carelessly creates duplicate counting.
GA4-imported conversions use GA4's attribution model (data-driven by default), which can differ from Google Ads' own click-based attribution for native tags. If you import a GA4 "purchase" event AND have a native Google Ads purchase conversion action both marked as "primary" for bidding, Smart Bidding may be optimizing against double-counted or inconsistently attributed data.
A clean setup picks one source of truth per conversion action: either native Google Ads tracking (simpler, click-attribution, easier to debug) or GA4 import (richer, cross-channel attribution, but requires the GA4-Google Ads link and consistent event configuration). Mark only the conversions you want Smart Bidding to optimize toward as "Primary" — set secondary/observational conversions to "Secondary" so they are visible in reports without influencing bids.
Auditing and verifying your conversion setup
A quarterly conversion audit — and an immediate one after any site redesign, platform migration (e.g., moving to a new e-commerce platform), or checkout flow change — catches problems before they distort months of bidding data.
Use Google Tag Assistant (or GTM Preview mode) to manually walk through each conversion path and confirm the tag fires exactly once, with the correct value, at the right point in the flow. Cross-check the conversion count in Google Ads against your actual order/lead volume from your CRM or e-commerce platform for the same period — a gap of more than a few percent in either direction signals a tracking issue.
- Walk through each conversion path manually using GTM Preview or Tag Assistant
- Compare Google Ads conversion counts to actual orders/leads from your CRM or order system for the same date range
- Check for duplicate tags firing on the same page (a common artifact of platform migrations where old and new tags both remain)
- Verify conversion values are dynamic and accurate for variable-value purchases, not hardcoded
- Confirm "Primary" conversion actions used for bidding are not double-counted across native tags and GA4 imports
Using Claude to audit conversion data
With MCP Ads connected to both Google Ads and GA4, Claude can cross-reference conversion counts between the two systems directly: "Compare the number of purchase conversions reported in Google Ads vs the purchase events in GA4 for the last 30 days, broken down by week. Flag any week where the two differ by more than 10%."
You can also ask Claude to scan for the most common setup mistakes: "List all active conversion actions in this account, show which are marked Primary, and flag if any look like duplicates — for example, both a native purchase conversion and a GA4-imported purchase conversion marked Primary at the same time."
Doing this with Claude connected to the account
Conversion tracking problems are hard to find because the evidence is split across three products: the conversion actions in Google Ads, the tags in Tag Manager, and the events in GA4. MCP Ads connects all three (Google Ads, GA4, and GTM), so Claude can check them against each other in one pass.
Prompt 1: "Audit my conversion setup. List every conversion action with its status, category, counting method, last conversion date, and whether it is set as primary. Then tell me if auto-tagging is on." Claude calls google_ads_list_conversion_actions_full and google_ads_check_auto_tagging, then summarizes with google_ads_conversion_setup_complete. What comes back is the list of problems rather than the list of actions. As an illustrative example: two primary actions counting every conversion, one page-view action that has not fired in 45 days, and auto-tagging disabled, which explains why GA4 is showing the traffic as generic referral.
Prompt 2: "Compare purchases in GA4 against the Google Ads purchase conversion action for the last 30 days, day by day, and show where they diverge." This uses ga4_conversions alongside google_ads_conversion_performance_by_action, and ga4_check_google_ads_integration to confirm the two properties are actually linked. If the gap is large, ga4_diagnose_missing_conversions checks the usual causes — key event not marked, consent mode dropping hits, or a tag firing on the wrong page.
Prompt 3: "Create a new conversion action for Contact Form Submit, counted once per click, and build the matching tag in GTM on the form-submit trigger, but do not publish the workspace." Claude uses google_ads_create_conversion_action, then gtm_setup_google_ads_conversion to build the tag and trigger in a workspace. Every one of those is a write, so Claude shows the conversion name, value settings, and the GTM container it will touch before doing anything, and the workspace stays unpublished until you explicitly ask for gtm_publish_workspace_now.
The Google Ads skills include a campaign health check that repeats the first prompt every morning and flags a tracking outage the day it happens rather than at month end.
Where the manual way still wins
Claude can read the configuration, but it cannot load your website in a browser and watch the tag fire. Tag Assistant, GTM preview mode, and the browser network tab remain the only way to confirm a tag actually sends a hit with the right value on a real form submission. The API also cannot see consent banner behaviour or whether a developer wrapped the thank-you page in a redirect that never renders.
What counts as a conversion, and what each one is worth, is a business decision. Claude will happily create a primary action for a newsletter signup; whether that should steer Smart Bidding is your call. Set the value rules, decide which actions are primary, and then let Claude verify that the setup matches what you decided.
