Short answer: treat a rise in branded search after paid social as a demand signal, not proof of causation. Start with a clean baseline and a lag-aware report, then move to a platform lift study or a matched geographic test when the decision is important enough to justify controlled measurement. Reconcile paid, organic and direct outcomes in one table, and keep attribution, incrementality and marketing-mix modelling as separate questions.
Paid social can influence a person before that person searches for a brand, while a search platform may receive the final click. A last-click or short attribution window can describe where a conversion was captured without answering whether the upstream social exposure changed the outcome. The reverse risk is just as real: a branded search can reflect demand that already existed. The job is to estimate the counterfactual—what would have happened without the social treatment—before changing budgets.
What the evidence ladder should answer
Use the least expensive method that answers the decision in front of you. An observational dashboard can tell you whether branded searches and social exposure move together. It cannot, by itself, rule out seasonality, promotions, competitor activity, distribution changes or an already-growing brand. A controlled lift design can estimate a causal difference, but it requires eligibility, volume, clean controls and enough time to finish. A marketing-mix model can provide a portfolio view, but its counterfactual remains dependent on data quality and model assumptions.
- Observe: establish a branded-search baseline and inspect timing.
- Explain: map the real conversion lag and isolate contamination.
- Test: use Search Lift, Conversion Lift or a geo holdout when the question is causal.
- Plan: use a model such as Meridian only with suitable geo-time data and transparent assumptions.
1. Build a branded-search baseline
Define brand, product and misspelling query groups before looking at the result. Pull daily impressions, clicks, conversions and cost for paid search, plus Search Console clicks and impressions for matching query groups where the data is available. Keep query definitions stable. Record social spend, impressions, reach, frequency, campaign objective, creative change dates and geo coverage beside those search fields.
A 30-to-60-day window can be a practical starting point for an observational baseline, as the trade discussion that prompted this guide suggests, but it is not a universal sample-size rule. Choose a window long enough to show ordinary weekly patterns and short enough that the media plan, brand, prices and distribution have not changed beyond recognition. Google Search Console has data-freshness and anonymisation limitations, so use complete days, label preliminary data and avoid treating a missing query row as zero demand. The Search Console performance report, data-freshness guidance and dimensions and limitations guidance are the control points for those filters.
During the observation period, document what stayed constant. If paid-search budget, bids, match types, promotions, landing pages or competitor activity changed at the same time as social spend, the comparison is confounded. A corresponding rise in branded queries after a social spike is worth investigating; it is not a lift estimate. Report the direction and date range with known changes attached, not as a promise about future performance.
2. Add lag without inventing a lag
Do not copy a 14-day lag from another advertiser. Compute a candidate lag from your own customer journey: first exposure or click, search, lead, opportunity and revenue timestamps. Use cohorts where possible, and show the distribution rather than only an average. For a long sales cycle, a same-day ROAS view may be incomplete. For a short cycle, a long lag can create false associations.
Keep lag analysis descriptive until a test supports a causal conclusion. Plot social exposure and branded-search demand at several reasonable lags, mark promotions and outages, and report whether the relationship survives ordinary control variables. Avoid selecting the lag that produces the most flattering chart. If the business question is about incremental conversions, use a controlled design or a model that makes its counterfactual explicit.
3. Use platform lift when it fits the question
Google describes Search Lift as a treatment/control measurement of whether ads change search behaviour. Its setup guidance says availability depends on the account and describes limits for search terms and groups. Check the About Search Lift and setup documentation in the actual account before promising a study. A Search Lift result answers a search-behaviour question; it is not automatically proof of revenue or of a particular Meta campaign’s effect.
Google’s broader lift-study guidance separates Search Lift from Conversion Lift. Conversion Lift asks about incremental conversions or value, with treatment and control groups. Its geo-based reporting explains that conversions are aggregated into non-overlapping regions and that output can include incremental conversions, value, cost, iROAS and confidence intervals. Read the final study result and any insufficient-data or no-lift status; an early result is not a budget recommendation. The geo-based reporting documentation describes the relevant metrics and caveats.
Meta provides first-party training on Conversion Lift and on search-lift measurement. Those materials are useful for understanding the test concept and data setup, but exact account eligibility, partner requirements and UI availability must be confirmed in the account. Do not imply that a Meta study measures Google’s search inventory unless the actual study design and outcome definition say so.
4. Design a matched geo test when the decision is causal
A paired geo design keeps the business question concrete: does changing paid social in treatment regions change search or business outcomes relative to comparable control regions? Pre-register the regions, treatment, control, pre-period, test period, primary outcome, guardrails, exclusion rules and stopping policy. Keep paid-search settings and non-social activity as comparable as practical. Log contamination, such as a national promotion, competitor price change, delivery restriction or a campaign that leaks into control.
Use a pre-period to test whether regions move together before treatment. Choose the outcome before opening the post-period report: branded-search impressions, total search conversions, qualified leads, revenue or a defensible micro-conversion. If revenue takes longer to arrive than the test, define a cooldown and label the result incomplete until the window closes. Do not publish a percentage lift, confidence interval or iROAS unless it came from the actual experiment and the method supports that calculation.
Google’s GeoX analysis guidance describes counterfactual time-series reasoning, while its GeoX FAQs explain why geo experiments can compare channels but still demand careful design. The Meridian data-collection guidance is a checklist: media and outcome data need consistent geography and time keys, and control variables need to be recorded rather than reconstructed from memory.
Method selection table
| Method | Question it answers | What to record | Main limitation |
|---|---|---|---|
| Baseline and lag report | Did search demand move after social exposure? | Query groups, spend, exposure, dates, lag candidates and confounders | Association; cannot isolate causation |
| Search Lift | Did eligible advertising change search behaviour? | Study eligibility, treatment/control, terms, lift metric and confidence | Account thresholds and search outcome do not equal revenue |
| Conversion Lift | Did the tested campaign change conversions or value? | Test design, outcome definition, incremental result, interval and final status | Requires sufficient volume and a valid experiment |
| Matched geo holdout | Did changing social cause a downstream regional difference? | Geo assignment, pre-period fit, treatment dose, controls, leakage and cooldown | Noise, cost and contamination can weaken the read |
| MMM or Meridian | How does social contribute within the whole media portfolio? | Summable geo-time media, outcomes, controls, priors and diagnostics | Counterfactual depends on data and model assumptions |
Reconcile the result before changing budget
Create one row per date, geo and channel where the data permits. Keep separate columns for social spend and exposure, branded paid-search impressions, clicks, conversions, non-brand search outcomes, organic Search Console observations, direct outcomes, qualified CRM outcomes and revenue. Add campaign objective, creative version, bid or budget change, promotion flag, competitor event and data-freshness status. The row should make it possible to explain why a metric moved without silently assigning every downstream conversion to the last click.
Start with the broad Google Ads measurement stack as the parent framework, then keep this cross-channel question in its own report. A platform metric can remain useful for optimisation while a lift result calibrates the business decision. For mobile or app measurement, the cross-channel attribution setup shows why event definitions and partner boundaries matter. For video, distinguish YouTube views versus engaged views before comparing exposure across channels.
Operational context matters too. The AI for Google Ads guide and the Demand Gen guide cover campaign surfaces that may alter delivery. Keep those changes in the experiment log. The Google Ads bidding checklist is a reminder that a bid-strategy change can look like channel lift if it is not recorded. A durable martech stack guide can help assign data owners, but tooling does not remove the need for a defined counterfactual.
What not to claim
- Do not call a branded-search spike incremental without a control, lift study or clearly stated model.
- Do not treat agency case-study percentages as benchmarks for a different advertiser.
- Do not backfill missing Search Console or CRM rows with zeroes.
- Do not mix a platform-attributed conversion, an incremental conversion and a qualified lead in one KPI.
- Do not move budget while a test is still collecting its defined conversion lag or cooldown.
FAQ
Is branded-search growth enough to prove paid social worked?
No. It is a useful diagnostic because it can reveal a change in declared demand, but seasonality, promotions, public relations, competitor activity and existing brand momentum can produce the same movement. Use it to decide whether a controlled test is warranted.
Should paid search be paused in the control region?
Not by default. The test should match the business question. If the question is whether social creates demand that search captures, document the paid-search treatment and keep it stable as practical. If the question is the incrementality of brand search itself, that is a different test with different risk and guardrails.
Which method should a smaller advertiser use?
Begin with a defined baseline, complete-day data and a pre-registered pre/post or time-based analysis. A lift study or geo test is appropriate only when the account has the required eligibility and enough outcome volume. The modern-measurement playbook recommends matching the method to the decision rather than adding an experiment for its own sake.
Where does MMM fit?
MMM is useful for portfolio allocation and longer horizons when media, outcomes and controls are consistently collected. It should complement, not erase, a lift result. Use model assumptions and uncertainty to explain a range of plausible contributions, not to manufacture a precise causal number from sparse data.
Bottom line
Paid social may create demand that paid search later captures, but the credible measurement path is staged: observe branded search, respect the actual lag, test incrementality when the decision matters, and reconcile the result with business outcomes. Preserve the difference between a signal and a causal estimate. That discipline produces a budget decision a finance or growth team can audit, even when the final answer is that the evidence is not yet strong enough to scale.
