How to Use Claude for Google Ads: From Brief to Review

To use Claude for Google Ads, give it an approved campaign brief and relevant exports, then ask it to group keywords, draft ads, and explain performance changes. Review its evidence and approve each proposed change before applying it in Google Ads. Start with one Search campaign so you can check every recommendation against the offer and landing page.

Tools like fenbs keep campaign rules, reusable prompts, and experiment decisions outside a single chat, so the next session starts with the same approved instructions.

How do you start the workflow in Claude?

1. Write the goal: name the offer, audience, service area, landing page, and action you want visitors to take.
2. Prepare the evidence: supply approved business facts and relevant campaign or keyword exports, with their dates and filters.
3. Ask for keyword groups and ad drafts separately. Review the groups before generating copy for them.
4. Have a person check claims, targeting, spending settings, and measurement before making changes in Google Ads.
5. Return with comparable performance data. Ask for observations, uncertainties, and one proposed experiment rather than a list of immediate edits.

You can start with an ordinary conversation; our Claude introduction explains the basics. This file-based workflow does not require giving Claude access to your advertising account. Account-changing connections are a separate topic in AI assistants for Google Ads.

What belongs in the campaign brief?

Describe the business as it actually operates. Include the landing page text, approved claims, geographic coverage, unavailable services, brand voice, and who approves changes. Define a conversion, such as a qualified appointment request, so Claude does not treat every click or form submission as equally valuable.

Say an Austin roofing company wants roof-repair inquiries. Its brief should distinguish repair work from full replacements, identify the areas it serves, and state whether emergency appointments are available. Those details stop a plausible draft from advertising a service the company cannot deliver.

Remove customer details and anything you lack permission to share. Claude supports CSV uploads; check Anthropic’s file upload guidance↗ (opens in a new tab) for supported formats and current limits. Give each export a clear name, reporting period, currency, and explanation of its columns.

Keep standing instructions, such as requiring approval for new claims, in fenbs Decisions and rules, so another assistant can read them before drafting.

How do you group keywords without inventing demand?

Use measured inputs. For a new campaign, get ideas and estimates from Google Keyword Planner↗ (opens in a new tab). Ask Claude to organize that export by search intent, the reason behind a search, and by the landing page that could answer it. Generated keyword ideas are suggestions to validate, not evidence of demand.

For an existing campaign, download the Google Ads search terms report↗ (opens in a new tab). A search term is what someone searched; a keyword is what you target. Keep those columns separate, along with the campaign, ad group, and match type. Google omits some low-activity queries for privacy, so this report is not a complete list of searches.

• Group relevant searches by the service requested and the destination page, with a short reason for each grouping.
• Separate unrelated searches, such as roofing jobs, from potential customers looking for roof repairs.
• Mark uncertain terms for review. Do not turn every unfamiliar phrase into a negative keyword, which excludes matching searches.
• Return the original row identifier beside every suggestion so a person can trace it back to the export.

How do you turn approved groups into ad drafts?

Choose one approved group and its landing page. Ask for distinct messages about the service, a supported benefit, and the next action. Require Claude to show the supporting business fact beside each claim, and to flag missing evidence instead of filling gaps with persuasive language.

For responsive search ads, Google combines headlines and descriptions in different orders. Each piece should make sense alone and with the others. Use Google’s responsive search ad documentation↗ (opens in a new tab) for current field limits, then check character counts in a spreadsheet or the ad editor rather than trusting Claude’s count.

Use only the attached campaign brief, approved keyword group, and landing page text.
Draft responsive search ad headlines and descriptions within the field limits I provide.
Return: asset type, draft text, supporting source, and questions for the reviewer.
Make each asset understandable on its own and compatible with the others.
Do not invent guarantees, credentials, availability, discounts, or results.
Do not publish or change account settings. Stop where evidence is missing.

What should a person check before publishing?

Read the draft beside the actual landing page. Check that the service exists, the destination works, the offer is current, and the contact method reaches the right team. Remove unsupported guarantees and credentials. Then inspect combinations of headlines and descriptions for repetition, contradictions, or promises whose qualifications could disappear.

Google holds advertisers responsible for following applicable laws and advertising policies, including content on destinations. Its policy enforcement guidance↗ (opens in a new tab) explains that disapproved ads cannot run until corrected and reviewed. Claude’s review is a useful extra pass; it cannot establish that your ad meets every requirement.

Before applying approved changes, check location settings, keyword exclusions, spending settings, and the conversion action being measured. Save the exact approved copy and the reviewer’s name. A draft labeled ready is not evidence that anyone approved publishing it.

How do you review performance without chasing noise?

Supply two comparable reporting periods with the same filters, conversion definitions, and currency. Include impressions, clicks, cost, conversions, and conversion value when relevant. Explain any tracking changes, promotions, or landing page edits. Ask Claude to check totals against the export before interpreting differences.

Request three separate outputs: observed changes with the supporting rows, possible explanations, and evidence needed to choose between them. An increase in clicks does not prove better business results. Recalculate rates from totals rather than averaging row percentages, and ask Claude to show its formulas.

If qualified inquiries are the goal, include a reviewed summary of lead quality. Do not assume that every conversion recorded in the advertising report represents a suitable customer.

Check whether recent conversions are still arriving. Google explains that conversion lag↗ (opens in a new tab), the delay between an ad interaction and a conversion, can make recent cost per conversion look higher than it eventually will. Ask Claude to flag that uncertainty before recommending a pause or spending change.

Choose one hypothesis to test, such as whether clearer service wording attracts more qualified inquiries. Record the proposed change, success measure, review date, and conditions for stopping. A person chooses the experiment and approves implementation; a small or incomplete sample can justify waiting for more evidence.

How do you keep useful prompts and decisions?

Keep the prompt that worked, the inputs it expects, and the checks a reviewer must perform. Record experiment choices separately from observations. “Repair wording was approved” is a decision; “repair inquiries increased” needs evidence and may still have several explanations.

A custom application for repeated requests is a separate project; our Claude API guide covers that route. First make the manual workflow reliable enough that another person can repeat it.

How to do this with fenbs

1. Create a free account, then add a project for the campaign on your kanban board.
2. Save the reusable drafting prompt in Skills and prompts. Record the owner’s approved claims and standing approval requirements in Decisions and rules.
3. Add a task for the first experiment. Put the hypothesis, evidence, proposed change, and checks in its plan.
4. Connect Claude using the integration guide. Ask it to read the project context before working and save progress before stopping.

fenbs is free for one person, with no credit card. It does not connect to Google Ads or publish campaigns. If you only need one ad draft, a reviewed document may be enough; use fenbs when prompts and decisions need to survive across sessions. Explore more workflows in AI at work.