Perspective / Search & performance marketing
Migrating to Google AI Max: conversion signals, brand controls and controlled rollout

A Google AI Max migration should be treated as a controlled change to a commercial decision system. Matching, generated text and landing-page selection can change together, while bidding continues learning from the conversion signals supplied to it. Success therefore depends on more than activating a setting. The practical work is to verify the goal, define acceptable traffic and claims, configure the available controls and compare mature commercial outcomes. This article sets out a migration framework for a hypothetical B2B software provider, rather than promising a universal performance uplift.
The executive view
- Confirm the current migration timetable and the specific campaigns it affects.
- Validate conversion quality before expanding how campaigns find and serve traffic.
- Judge rollout using mature commercial outcomes alongside explicit operational guardrails.
Establish which change and timetable actually apply
Begin with an inventory of Search campaigns, existing Dynamic Search Ads, automatically created assets, broad-match settings and landing-page rules. Record which capabilities are already active so that a later result is not attributed to something the account was doing before the migration.
Google’s migration announcement was published in April 2026 and updated on 11 June. The explicit update moves the start of automatic Dynamic Search Ads upgrades to February 2027, while retaining a September 2026 timetable for specified campaigns using automatically created assets and campaign-level broad match. Distinguish these cases; older wording within the same announcement can otherwise lead to an incorrect DSA deadline.
Before implementing, confirm the current notice and settings visible in the actual account. A published transition plan is not evidence that every campaign has already changed. Preserve an export of the baseline configuration and name the person responsible for the migration decision.
Consider a hypothetical account containing a DSA acquisition campaign, a conventional Search campaign already using broad match, and a separate product campaign with automatically created assets. They may face different transition conditions even though they share one billing account and conversion action. Map each proposed setting change to the campaigns it actually affects. A shared conversion-goal edit can alter the comparison group too, so check inherited settings before describing that group as unchanged. Sequence work around these dependencies and record which campaign can provide a useful baseline while another changes.
Check what the campaign is being asked to optimise
A campaign optimising for every form submission may efficiently generate enquiries that sales cannot qualify. Audit the conversion actions included in bidding, duplicate events, imported CRM stages, conversion windows and value definitions. Separate a genuine customer outcome from a measurement event that happens to be easy to collect.
For the hypothetical software provider, an introductory download is a useful diagnostic event, a sales-accepted opportunity is a stronger quality signal and a signed contract is a later commercial outcome. The choice of bidding goal depends on signal volume and delay as well as economic relevance. Do not switch all campaigns to a sparse late-stage event without assessing whether sufficient feedback exists.
Write down how conversion value is calculated. An unweighted multi-year contract headline, first-year revenue and expected contribution are different quantities. Consistent definitions matter more than attaching a superficially sophisticated value to every lead. Review the CRM reconciliation before changing the traffic system that learns from those values.
Audit the route from the originating lead to the reported conversion. Can a submitted form be classified consistently as a relevant organisation, a qualified requirement and an eventual commercial outcome? Do imports preserve occurrence time, currency and stable event identity? A change in CRM integration can alter campaign learning at the same moment as matching expands. Freeze or document such changes during evaluation. If sales rejects spam and duplicate enquiries manually, connect those outcomes to the analysis before declaring improved efficiency. More low-cost form submissions can increase the workload of the very team needed to convert genuine demand.
Translate the feature set into commercial boundaries
Google’s AI Max overview describes search term matching, text customisation and final URL expansion. These capabilities affect different parts of the journey, so review them separately even when they are offered within one suite.
For search matching, identify out-of-scope products, recruitment queries, support intent and geographic constraints. For generated text, identify regulated or unsupported statements, contractual promises and terminology the business must use precisely. For URL expansion, decide which pages can accept paid traffic and which should be excluded from that role.
An account-level assumption such as “all indexed pages are suitable landing pages” is weak for a technical B2B site. A documentation page, legacy product page or article may be relevant to a query but poor at setting the right commercial expectation. Inspect the controls available for the particular campaign and record both what they enforce and what still needs monitoring.
Treat individual capabilities as separately understood delegations even when they are enabled through a common product setting. Expanded matching changes which searches are eligible. Generated text changes how the offer is expressed. URL expansion changes which destination can receive the visit. Each can create a different failure mode, requiring a different check. Google's AI Max operation guide describes feature behaviour and controls; verify the live configuration rather than assuming that a brand list or URL rule restricts every feature in the same way. Document the practical reach of each setting.
Use a boundary register and a landing-page contract
Create a concise register that joins each commercial constraint to its implementation and verification method. This prevents the migration becoming a collection of settings whose business purpose is forgotten.
| Illustrative boundary | Implementation question | Evidence to review |
|---|---|---|
| Sell the current CRM product only | Which matching and exclusion controls apply? | Query and conversion-quality samples |
| Avoid unsupported delivery promises | What generated-text guidance is available? | Live text and destination review |
| Send buyers to approved commercial pages | How are URL expansion and exclusions configured? | Actual landing-page reporting |
| Evaluate UK-qualified opportunities | Which geographic and CRM rules are applied? | Location and sales-stage reconciliation |
Each approved landing page also needs a contract: intended audience, supported offer, accurate claims, measurable primary action and owner. Fix contradictions in the website before allowing broader automated selection. Campaign control cannot compensate reliably for an ambiguous product catalogue or outdated commercial content.
Test exclusions with representative edge cases. An approved commercial page may link to a careers section, an outdated offer or a technical document unsuitable for acquisition traffic. Decide whether URL rules operate on exact pages, prefixes or broader website areas, and verify the result in destination reporting. Review redirect targets and mobile destinations, not just the URL entered in a configuration screen. For brand controls, retain examples of allowed and disallowed search intent. A control register becomes meaningful when actual traffic can be checked against these concrete examples by a named reviewer.
Design a comparison that answers the rollout question
Define the question before launching the test. Is the objective to expand useful reach, reduce the cost of qualified pipeline or maintain quality while saving management time? Choose one primary commercial measure and a small set of guardrails. Avoid selecting whichever metric improves after the fact.
Use the platform’s supported experiment mechanisms where appropriate and document the comparison conditions. Keep the offer, budget policy, conversion definitions and sales qualification rules stable enough to interpret the result. If several components change simultaneously, the test can evaluate the combined configuration but cannot cleanly assign credit to each feature.
Respect conversion delay. For a long sales cycle, early form volume can be reviewed as an operational signal while qualified opportunities and sales mature. Establish the evaluation period and minimum evidence needed in advance. A few days of favourable platform reporting is a weak basis for reallocating a material commercial budget.
Estimate whether the account can produce an interpretable result before running the comparison. If only a handful of qualified opportunities arise each month, a short split test can leave uncertainty larger than the decision threshold. Consider a longer observation period, a larger suitable scope or a staged operational assessment, while stating the causal limitations. Avoid interrupting the test for ordinary day-to-day variation. Equally, do not wait for statistical certainty when a hard boundary such as a prohibited claim is breached. Commercial evidence thresholds and immediate operational stop conditions answer different questions.
Account structure can also undermine comparability. If the baseline and candidate compete for the same eligible traffic, or adjacent campaigns change brand coverage during the test, the result may reflect redistribution within the account. Record concurrent promotions, pricing changes and sales-capacity constraints that alter the conversion path. Use the platform's supported experiment design and explain residual overlap or interference. An experiment report should state the scope of the conclusion: performance under these account conditions, not a universal forecast for another market, product group or season.
Work through a result with quality and margin included
Suppose a baseline campaign spends £20,000 and produces 100 enquiries, of which 30 become sales-accepted opportunities. A candidate configuration spends the same amount and produces 130 enquiries, with 26 accepted opportunities. Cost per enquiry improves from £200 to about £154, while cost per accepted opportunity worsens from about £667 to £769.
That does not prove the candidate is commercially inferior: opportunity value, close probability and later revenue may differ. It does establish that enquiry efficiency alone cannot justify rollout. Inspect whether broader matching changed the product mix, whether sales applied the same qualification rules and whether delayed outcomes have matured equally.
Use this decomposition to guide investigation rather than attributing every movement to the AI setting. Measurement changes, seasonality, competitors and sales capacity can all affect the comparison. Vendor aggregate uplift figures describe vendor-reported samples; they are not a forecast for this account or a substitute for its own evidence.
Extend the illustrative comparison to expected contribution. If the baseline's thirty opportunities each have a 25 per cent close probability and £8,000 expected contribution when won, their expected contribution is £60,000. If the candidate's twenty-six opportunities have the same probability and £10,000 contribution, the figure is £65,000. After the same £20,000 spend, the estimates become £40,000 and £45,000 before sales handling costs. These are assumptions, not measured profit or a forecast. Small changes in close rates can reverse the ranking. Show sensitivity and later reconcile the assumptions with mature cohort outcomes.
Separate platform optimisation from incremental demand
Even a well-run campaign comparison does not automatically establish how much new demand advertising created. A configuration can receive more attributed conversions by capturing buyers who would otherwise have returned through another route. Distinguish the operational question of which setting performs better within the account from the investment question of whether the advertising creates sufficient additional contribution.
A hypothetical management report can show three layers: platform-reported outcomes for operational feedback, CRM cohort progression for commercial quality, and a separate incrementality assessment where the business can support an appropriate design. Keep the definitions visible. A rise in attributed revenue alongside unchanged total qualified pipeline calls for investigation, especially if brand capture increased. It is not proof of failure, but it limits the conclusion supported by attribution alone.
Plan the next decision rather than treating migration completion as the final milestone. The sponsor might approve one business unit or product group, subject to destination compliance, lead-quality stability and agreed spend limits. Wider deployment then depends on the evidence from that scope. This makes adoption a sequence of commercial decisions with traceable assumptions, rather than a single irreversible judgement about whether an AI-labelled product is good or bad.
Release with monitoring and a defined response
Agree what happens when a guardrail is breached. A disallowed claim may require immediate intervention; a modest shift in opportunity cost may justify further observation. Name the owner, response time and permitted corrective actions so that monitoring can change behaviour.
Preserve the ability to compare configurations, but do not imply that every platform transition can be reversed indefinitely. Distinguish an experiment rollback from a permanent product migration imposed by the platform. Record the current options and keep exports of settings, approved destinations and measurement definitions.
For the executive sponsor, the approval is a bounded delegation: this campaign may use these capabilities, pursue these commercial outcomes and operate within these constraints. Expand adoption when evidence supports that delegation. The technical achievement is a traceable connection between automation, business rules and realised outcomes, rather than an account-wide switch justified by enthusiasm for a new feature.
Keep an explicit audit trail when changing budget, goal configuration or commercial rules. Record what changed, who approved it and which evaluation period it affects. Review search terms, actual destinations, lead rejection reasons and pipeline cohorts together, rather than delegating separate dashboards to teams that never reconcile them. A practical release gate requires a validated measurement path, understood controls, an owner for exceptions and enough mature evidence to justify the next scope increase. When platform functionality changes again, this operating discipline remains useful even if the particular migration timetable no longer applies.
Questions this raises
When are Dynamic Search Ads being upgraded to AI Max?
Google’s 11 June 2026 update says automatic DSA upgrades will begin in February 2027. Other campaign settings have different transition dates, so check current account notices before planning the change.
Should AI Max be judged by cost per lead?
Cost per lead is useful but incomplete. Compare lead quality, opportunity value, conversion delay and commercial outcomes using consistent definitions across the baseline and candidate configurations.
Can a campaign experiment identify which AI Max feature caused an improvement?
Only if the design isolates that feature adequately. Changing matching, text and destinations together evaluates the combined configuration; it does not establish each component’s individual contribution.
Sources & further reading
- We’re upgrading Dynamic Search Ads to AI Max 15 April 2026; updated 11 June 2026
- About AI Max for Search campaigns Ongoing guidance; checked 17 November 2026
- How AI Max for Search campaigns works
From thinking to doing