How to Create Google Ads with AI: A Precision Guide

What AI should and should not do in the build

AI tools are good at three parts of building Google Ads: grouping hundreds of search phrases by intent, drafting many headline and description options within character limits, and brainstorming the searches you do not want. They are poor at deciding what you can truthfully claim, what a lead is worth, and how much to spend. This guide keeps those decisions with a person and gives the rest to AI.

Step 1: define the conversion before anything else

Decide what the campaign is for: a quote request, a booked call, a purchase. Set that as the primary conversion and assign it a value, even an estimate based on your average deal. If sales happen offline, plan to import qualified outcomes using enhanced conversions for leads (Google Ads Help). Everything after this step optimises towards this choice.

Step 2: gather real search language

Collect phrases from sources that reflect how buyers speak:

• • Google Keyword Planner ideas for your services and location.
• • The search terms report from any existing campaigns.
• • Words customers use in enquiry emails, call notes and reviews.
• • Competitor and category terms you would genuinely want to appear for.

Paste the list into your AI tool and ask it to cluster by intent, not by shared words. "Office cleaning prices" (comparing costs) and "office cleaning jobs" (looking for work) share words and nothing else.

Step 3: structure the campaign

• • One campaign per goal and budget: for example, a brand campaign and a non-brand services campaign.
• • One ad group per intent cluster, small enough that one set of headlines is relevant to every keyword in it.
• • Match types: phrase and exact match give tighter control; broad match relies on Smart Bidding and good conversion data. If AI Max is available or applied to your campaign, decide deliberately whether its search term matching and final URL expansion are on, and review the effects weekly.
• • Location set to people in or regularly in your area, if you serve a local market.

Step 4: write the responsive search ad to a slot plan

A responsive search ad takes up to 15 headlines of up to 30 characters and up to 4 descriptions of up to 90 characters (Google Ads Help). Google assembles combinations, so plan headlines by role so that any combination still works.

Illustrative example: a hypothetical commercial cleaning company in Manchester, ad group "office cleaning".

Illustrative headline slot plan for an office cleaning ad group (character counts included)

Role	Headline	Characters	Why it is there
Keyword echo	Office Cleaning Manchester	26	Confirms the search was understood; supports ad relevance
Keyword variant	Commercial Cleaners Manchester	30	Covers the other common phrasing
Main benefit	Cleaned Before Staff Arrive	27	The outcome the buyer cares about
Proof	Fully Insured and DBS Checked	29	Reduces the risk of letting strangers into the office
Offer	Free Site Visit and Quote	25	A low-commitment next step
Scope	Daily, Weekly or One-Off	24	Answers 'do they do what I need?'
Objection	Same Cleaner Every Visit	24	Answers a common worry about turnover
Call to action	Request Your Quote Today	24	Tells the searcher what to do

Descriptions carry what headlines cannot. The illustrative set for the same ad group:

1. • "Evening and early-morning office cleaning across Manchester, planned around your hours." (87 characters)
2. • "Insured, DBS-checked cleaners. Free site visit, then a fixed quote for your schedule." (85)
3. • "Daily, weekly or one-off cleans for offices, clinics and shared workspaces." (75)
4. • "Tell us your hours and floor space. We visit, then send a fixed monthly price." (78)

Prompting tip: give the AI the ad group's keywords, the facts you can prove, the roles above and the limits, and ask for three options per role with character counts. Check the counts with a spreadsheet formula such as LEN, because models miscount. Delete any option that adds a claim you did not supply.

Pinning: avoid it unless something must appear. Pinning reduces the combinations Google can test and can lower Ad Strength. Ad Strength is a writing aid, not an auction factor; see why Ad Strength shows Poor.

Step 5: add assets

Add sitelinks to the pages a buyer would want next (pricing, sectors served, about, contact), callouts for short proof points, structured snippets for service types, and a call asset if you take phone enquiries. Character limits for each asset type are published in Google Ads Help; check them when you build rather than relying on memory.

Step 6: build the negative keyword list

Ask the AI to brainstorm searches that share your words but not your intent. For the cleaning example: jobs, vacancies, salary, training, courses, franchise, supplies, equipment, DIY, "how to clean". Review the list, then add it as a shared negative list. Revisit it weekly using the search terms report.

Step 7: connect your audience data

Upload customer lists through Customer Match, collected in a first-party context as Google's policy requires (Google Ads Help). Use them as observation audiences or exclusions (for example, to keep existing clients out of acquisition campaigns).

Step 8: pre-launch checks

• • Conversion action fires once per real conversion; test it.
• • Each ad group's final URL is the page that answers that intent, not the homepage.
• • Every headline makes sense alone and beside any other.
• • No claim appears that is not true and current.
• • Negatives are attached; location and schedule are right.
• • Budget is capped at a level you are prepared to spend while learning.

Step 9: the first 30 days

Check search terms twice a week in the first fortnight, then weekly. Do not change bidding targets in the first couple of weeks unless something is clearly broken. After 30 days, read the Quality Score components per keyword (Google Ads Help), which are a diagnostic of relevance, and the conversion data, which is the verdict. If a component is below average, the Quality Score guide explains what to change.

FAQ

Can I let Google's AI write the ads instead?

Google can generate text from your website, and AI Max includes text customisation. It saves time but does not know your banned claims or what is current. Many advertisers write their own core assets and review anything generated before relying on it.

How many ad groups should a new campaign have?

One per distinct intent you can write a specific ad and landing page for. A small local business might start with three to five; more than that on a small budget spreads data too thin.

Should I start with broad match?

Broad match works best when conversion tracking is reliable and the account already has conversion history. For a new campaign with little data, phrase and exact match give more control while Smart Bidding learns.

Which AI tool should I use for keyword clustering?

Any capable general-purpose model can cluster a keyword list if you instruct it to group by the searcher's intent and to explain each group. Check the groups by hand; the value is in the speed of the first draft.