Skip to content
Back to search

google-ads-brand-incrementality

"Measure how much paid Google search adds beyond organic for brand terms, or for any query family the site already ranks for, by reading Google Ads and Google Search Console data for the same queries together. Separates real demand from paid delivery, measures how much traffic organic recaptures after a brand budget, bid, pause or geography change, reads planned on/off and spend-shock tests, and designs a holdout when the evidence is not decisive. Use when someone asks whether brand spend is incremental, whether to cut, pause or raise brand, why brand clicks or conversions fell, whether organic would pick up brand traffic, or whether to bid on terms the site already ranks first for. Read-only: it never changes the account. Not for competitor campaigns or nonbrand terms the site does not rank for."

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/treboutat--google-ads-brand-incrementality/badge.svg)](/listings/treboutat--google-ads-brand-incrementality)

HTML

<a href="/listings/treboutat--google-ads-brand-incrementality"><img src="/api/listings/treboutat--google-ads-brand-incrementality/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

**As a Claude Code skill:**

mkdir -p ~/.claude/skills/google-ads-brand-incrementality
cp SKILL.md ~/.claude/skills/google-ads-brand-incrementality/SKILL.md
cp -r references scripts ~/.claude/skills/google-ads-brand-incrementality/

Then say "is our brand spend incremental?" and share the data below. The script needs only Python 3.

**In Claude.ai:** zip this folder and upload it as a custom skill in Claude's skills settings, or add `SKILL.md` and `references/method-rules.md` to a project. A project can't run the script, so use Claude Code or an upload with code execution for the numbers.

**As a plain SOP:** read `SKILL.md` and `references/method-rules.md`. They are a complete procedure even if you never run them through Claude.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

SKILL.md · Manifest · 132 lines
  1. ---
  2. name: google-ads-brand-incrementality
  3. description: "Measure how much paid Google search adds beyond organic for brand terms, or for any query family the site already ranks for, by reading Google Ads and Google Search Console data for the same queries together. Separates real demand from paid delivery, measures how much traffic organic recaptures after a brand budget, bid, pause or geography change, reads planned on/off and spend-shock tests, and designs a holdout when the evidence is not decisive. Use when someone asks whether brand spend is incremental, whether to cut, pause or raise brand, why brand clicks or conversions fell, whether organic would pick up brand traffic, or whether to bid on terms the site already ranks first for. Read-only: it never changes the account. Not for competitor campaigns or nonbrand terms the site does not rank for."
  4. metadata:
  5. author: TNT Growth
  6. version: "1.0.0"
  7. method-release: "2026-10-14.1"
  8. ---
  9. # Google Ads Brand Incrementality
  10. Paid brand clicks are not brand demand, and attributed brand conversions are not additional customers. This skill reads paid and organic results for the same queries together, so a change to brand spend is judged on what the business actually gained or lost. Nothing is changed in the account.
  11. ## 0. Load the method
  12. This file is the workflow. [references/method-rules.md](references/method-rules.md) holds the judgment rules (TNT Growth's Google Ads method, release 2026-10-14.1); read it before drawing conclusions. `scripts/brand_incrementality.py` does the arithmetic (Python 3, no packages). If a reference is missing, say so and do not claim it was applied.
  13. Route neighbouring requests elsewhere:
  14. - Competitor campaigns, or nonbrand terms the site does not rank for: no organic counterpart to read. Out of scope.
  15. - "Why did the whole account drop?": separate brand from nonbrand first, then use this skill for the brand part only.
  16. - Building or changing campaigns: out of scope. This skill recommends; a person with account access decides and acts.
  17. ## The idea in one table
  18. | Week | Paid clicks | Organic clicks | Total |
  19. |---|---:|---:|---:|
  20. | Ads on | 1,000 | 1,000 | 2,000 |
  21. | Ads off | 0 | 1,750 | 1,750 |
  22. Turning ads off lost 1,000 paid clicks, but organic picked up 750 of them. Only 250 clicks were truly additional: an incrementality ratio of 0.25. Everything below is this table with demand adjustment, noise bounds and checks.
  23. ## 1. Get the context
  24. Ask for, or find:
  25. - The business, where it can actually sell (states, countries), and what a real customer is.
  26. - Which campaigns serve brand (or the query family), and every change to them in the window: budgets, targets, pauses, **geography**, keywords and match types, networks, schedules.
  27. - The dates of promos, press coverage, launches, site or tracking changes.
  28. - Where business outcomes live (GA4, CRM) and how long a lead takes to become a customer.
  29. ## 2. Pull the data
  30. 1. **Paid, daily**, for the campaigns serving the query family:
  31. ```sql
  32. SELECT segments.date, campaign.name, metrics.impressions, metrics.clicks,
  33. metrics.cost_micros, metrics.conversions, metrics.search_impression_share
  34. FROM campaign
  35. WHERE campaign.id IN (<campaign ids>)
  36. AND segments.date BETWEEN '<start>' AND '<end>'
  37. ```
  38. Or a Google Ads UI export with Day, Impr., Clicks, Cost, Conversions and Search impr. share; the script reads both. Select campaigns by ID: a name filter like `LIKE '%Brand%'` also matches `NonBrand`. Check Performance Max and nonbrand for brand traffic too, since brand served there is invisible here.
  39. 2. **Paid by market.** Pull `geographic_view` with `segments.date` and `segments.geo_target_region` (impressions, clicks, cost, conversions) for both periods. Write two daily files: rows for the regions the business serves, and all other rows. Run the script on each. Paid geography often changes along with a budget change. Two limits: Search Console stays national (it filters by country only), so in-market paid is compared with national organic; and `geographic_view` has no impression share, so the second demand measure is not available for a by-market run.
  40. 3. **Organic, daily.** Search Console, Performance, Query filter set to a custom regex, Dates tab, export. Build the regex from the brand list and the paid search terms, then check the Queries tab and tighten it to navigational queries (for example `^(acme ?homes?(\.com)?|acme ?home login)$`). Loose filters pick up unrelated phrases that contain the brand, `site:` checks and AI-style questions. If Search Console is linked to Google Ads, `paid_organic_search_term_view` gives both sides by query.
  41. 4. **Business outcomes, daily, from all sources**: purchases, demos, qualified opportunities or signed customers from GA4 or the CRM, as a CSV with a `date` column and a `conversions` column (or name your column with `--outcomes-column bookings`). Give every lead the same window to convert (for example signed within 20 days of creation), or the newer period looks worse from lag alone.
  42. 5. **Auction Insights** for both periods (or at least search top and absolute-top impression share and rank-lost impression share).
  43. ### Demand measures
  44. The script reports two:
  45. - **Search Console impressions** for the filtered queries. The site usually ranks first for its own brand, so these track search volume. They may also move when what sits above the organic listing changes; treat that as a possibility to check, not a known effect.
  46. - **Paid impressions divided by search impression share.** This estimates eligible paid searches. It breaks when targeting, geography, keywords or match types change between periods.
  47. When they agree, trust the result more. When they disagree, find out why before reading any ratio.
  48. ## 3. Run the script
  49. ```bash
  50. # Before/after a change
  51. python3 scripts/brand_incrementality.py --paid paid.csv --organic gsc.csv \
  52. --change-date 2026-09-01 --pre-days 28 --post-days 28 \
  53. --exclude 2026-08-05:2026-08-07 --out report.md --json report.json
  54. # Planned test: 2x spend week, the normal weeks on both sides as baseline,
  55. # a promo week excluded, CRM outcomes
  56. python3 scripts/brand_incrementality.py --paid paid.csv --organic gsc.csv --outcomes crm.csv \
  57. --baseline 2024-05-27:2024-06-16,2024-07-01:2024-07-14 --treatment 2024-06-24:2024-06-30 \
  58. --exclude 2024-06-17:2024-06-23
  59. ```
  60. Match the windows to the real periods: if the change ran for two months, use a two-month treatment window (`--post-days 61`, or explicit `--baseline` and `--treatment` ranges), not the 28-day default. `--name "Category term"` relabels the report for a nonbrand family. `--gap-days` skips a transition after a change. `--organic-cvr` sets an organic conversion rate when there is no outcomes file. Run it separately per market, device or brand spelling when they behave differently.
  61. It calculates, per period, daily means of paid and organic impressions, clicks, cost and conversions, then:
  62. - **Demand change** on both measures.
  63. - **Expected clicks**: baseline paid and total clicks scaled by the demand change.
  64. - **Incrementality ratio** = (actual − expected total clicks) / (actual − expected paid clicks). 1.0 means every paid click lost was lost to the business; 0 means organic recaptured all of it. It works for increases and decreases.
  65. - **Cost per additional click**, and **cost per additional outcome** from the outcomes file.
  66. - **90% bootstrap interval** from resampling days within each period.
  67. Reading guide: 0.7 and above mostly additional; 0.3 to 0.7 partly additional; below 0.3 mostly recaptured. The report says the ratio is **not readable** when paid clicks barely moved after demand adjustment or the ratio falls far outside 0 to 1, and **unsettled** when the two demand measures disagree. In both cases it shows no cost per click. An interval that crosses zero is flagged: the data cannot rule out full recapture.
  68. ## 4. Judge the result
  69. 1. **Demand first.** Report the demand change before any paid number. Never describe a paid decline as a demand decline without it.
  70. 2. **Targeting before ratios.** Check whether geography, keywords, match types, schedule or networks changed between periods. A paid drop from narrower targeting is not lost demand or lost in-market coverage.
  71. 3. **Confounds.** The script flags short periods, small paid changes, demand swings over 30%, disagreeing demand measures, organic position shifts, paid impressions above Search Console impressions and wide intervals. Also check seasonality (the same weeks last year), promos, press spikes, SEO changes, tracking breaks and other channels (new social campaigns lift brand searches).
  72. 4. **Competitors.** Competitors entering brand auctions after a cut make paid brand more valuable than the click math shows, because they take the click and the customer.
  73. 5. **Outcomes, not clicks.** Decide on business outcomes by period. Split brand leads by the buyer's market, and compare the brand change with the whole business over the same months. Organic brand leads usually cannot be isolated in a CRM, because leads with no source mix organic, direct and app sign-ups; say so.
  74. 6. **Decision.** Price the additional outcomes against what the business can afford to pay for them. Keep brand's bid target based on its own mature cost. Do not cut brand because a ratio is low if competitors are present, and do not defend it on attributed conversions alone.
  75. ## 5. When to run a test
  76. Run one when there was no clean natural change, the demand measures disagree, the interval is wide, or the decision is budget-material.
  77. - **Geo, pre-post net of control:** run (or pause) the ads in test markets and not in matched control markets, and compare each group's change from its own pre-period. Search Console filters by country only, so a within-country readout needs GA4, product analytics or the CRM by region. Size it first; two weeks is usually too short.
  78. - **Time on/off or spend shock:** alternate weeks on and off, or double spend for a week, and read it with `--baseline` and `--treatment`, using the weeks on both sides as the baseline.
  79. - **Google's Conversion Lift**, where the account qualifies.
  80. Before the test starts, agree the primary outcome, the stop condition (for example a competitor taking top brand position, or an outcome drop beyond an agreed amount) and the owner. Any pause, exclusion or budget change is a live account change made by someone with authority to make it.
  81. ## 6. Output
  82. Lead with the answer: the demand change on both measures, then the ratio with its interval and reading, then the cost of the additional outcomes. Show the script's daily-averages table, the evidence limits, and what competitors did. Label a before/after result as observational, not as causal lift.
  83. ## Worked example (anonymized)
  84. A real estate brokerage's report, written by a general-purpose AI assistant, said a brand budget cut had hurt the account. It never looked at organic.
  85. - Search Console showed navigational brand searches up 9%. Paid brand clicks were down 23% and spend down 49%; total brand clicks were flat.
  86. - Read on all paid brand, the ratio was 0.33 (90% interval −0.05 to 0.55), which crosses zero and so was not a finding on its own. The second demand measure disagreed, which sent the analysis to targeting.
  87. - The geography split explained it. The brand campaign had been serving nationwide and was narrowed to the area the business serves. There, paid brand clicks rose while spend fell. The clicks that disappeared came from outside that area, and leads from outside it had never become customers.
  88. - In the CRM, brand leads from in-market buyers rose. Customers from them fell by about the same share as customers from all sources, so brand moved with the business, not against it.
  89. The budget cut saved money by dropping out-of-market brand ads, and the report's conclusion did not hold.
README.md · README · 49 lines
  1. # Google Ads Brand Incrementality
  2. A Claude skill that answers "is our brand spend actually adding anything?" It reads Google Ads and Google Search Console data for the same queries together, separates real brand demand from paid delivery, and measures how much of the traffic organic picks up when paid brand changes.
  3. It catches the classic misread: a paid brand drop that looks like collapsing demand, when organic quietly recaptured most of it, or when the ads simply stopped serving somewhere. It never changes the account.
  4. ## Who it's for
  5. Anyone deciding whether to cut, pause or raise brand spend, anyone asked "why did brand drop?", and anyone reviewing an analysis that blames or credits brand without looking at organic. It also works for nonbrand terms your site already ranks first for.
  6. ## Install
  7. **As a Claude Code skill:**
  8. ```bash
  9. mkdir -p ~/.claude/skills/google-ads-brand-incrementality
  10. cp SKILL.md ~/.claude/skills/google-ads-brand-incrementality/SKILL.md
  11. cp -r references scripts ~/.claude/skills/google-ads-brand-incrementality/
  12. ```
  13. Then say "is our brand spend incremental?" and share the data below. The script needs only Python 3.
  14. **In Claude.ai:** zip this folder and upload it as a custom skill in Claude's skills settings, or add `SKILL.md` and `references/method-rules.md` to a project. A project can't run the script, so use Claude Code or an upload with code execution for the numbers.
  15. **As a plain SOP:** read `SKILL.md` and `references/method-rules.md`. They are a complete procedure even if you never run them through Claude.
  16. ## What you provide
  17. 1. Daily Google Ads data for your brand campaigns: date, impressions, clicks, cost, conversions, search impression share (a UI export or a GAQL query).
  18. 2. The same split by region, so ads outside the area you serve can be separated.
  19. 3. Daily Search Console data filtered to your brand queries (Performance, custom regex, Dates tab).
  20. 4. The dates of brand changes, promos and press coverage.
  21. 5. Optional but best: daily business outcomes (purchases, demos, signed customers) from GA4 or your CRM.
  22. ## What you get
  23. - The demand change, read two ways, before any paid number.
  24. - The incrementality ratio with a 90% interval and a plain reading, or a clear "not readable" when there is no paid change to measure.
  25. - Cost per additional click and per additional outcome.
  26. - Evidence limits: short periods, demand swings, targeting changes, position shifts, filters that miss brand variants.
  27. - A test design (geo, on/off weeks or spend shock) when the before/after can't settle it.
  28. ## Files
  29. - `SKILL.md`: the workflow, with an anonymized worked example
  30. - `references/method-rules.md`: the judgment rules
  31. - `scripts/brand_incrementality.py` (Python 3, no packages): before/after windows, planned-test windows (`--baseline`/`--treatment`), outcomes files, date exclusions, two demand measures and a bootstrap interval. Reads Google Ads UI exports, GAQL output and Search Console exports.
  32. - `tests/`: a worked example and edge cases, plus unit tests for the script (`python3 -m unittest discover -s tests/scripts`)
  33. ## Related skills
  34. - [Google Ads Account Audit](https://github.com/treboutat/google-ads-account-audit): brand vs nonbrand separation is one of the checks.
  35. - [Google Ads Funnel Bottleneck Diagnostic](https://github.com/treboutat/google-ads-funnel-bottleneck): find which funnel stage broke before blaming brand.
  36. ---
  37. From [TNT Growth](https://tntgrowth.com/skills?utm_source=github&utm_campaign=treboutat). Use it, fork it, adapt it to your account.

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.