AI for Google Ads Client Questions: How to Answer Faster Without Guessing

This is how we answer client emails. Every step works by hand in the Google Ads UI and a spreadsheet, and AI takes over the tedious parts. We made up the store in the video above and generated its numbers, so no client data appears anywhere. The conversation and the math follow the same path we take on real accounts. The session in the video took about eleven minutes.

What AI does and what stays with a person

Most of the time on a client reply goes into mechanics: pulling several campaigns into one table, working out break-even ROAS, writing the email in plain language, and building a table that survives being pasted into an email. An AI assistant handles that:

• it turns the client's email into a list of questions and suggests what to ask,
• it pulls the campaigns and lays them out around the question,
• it works out break-even ROAS and the sale-price estimates,
• it builds a preview of each change before anything touches the account,
• it drafts the reply, and outside the video it also delivers it as an HTML file or a draft in the inbox.

Three decisions stay with a person: whether the questions to the client are the right ones, whether a change goes live, and whether the reply goes out. In the video, the account change waits for a yes, and the reply ends as a draft to read.

Step 1: Read the email as a list of questions

Every sentence in a client email carries a question the client didn't ask. "The sale starts Friday" asks when prices change on the site and in the product feed, because an ad showing the old price contradicts the landing page. "Make sure the ads push it" asks at what profit. "Budget is not a problem" isn't a number, and without a number you can't plan how much to add or when to stop.

What's missing matters just as much. Clients almost never mention margin or stock levels, and those two decide whether extra sales are profitable and whether the store can ship them.

Step 2: Ask before you look at the data

We ask three questions at most, one line each, so the client can answer in a minute. Here's what we asked:

• What's the margin on tents and sleeping bags after the 25% discount?
• Will the sale prices be in the product feed, and from what date?
• What's the cap on additional spend for the ten days?

The client answered that margin drops from 40% to 25%, sale prices go live on the site and in the feed Friday morning, and the cap on additional spend is $1,500.

The first question is always about margin. Without it you don't know where break-even sits, and every number after that is a guess. While you're on margin, confirm what it covers (shipping, payment fees, returns) and whether conversion value in Google Ads includes tax. If reported value includes sales tax and margin is calculated on the pre-tax price, multiply break-even by (1 + tax rate) before comparing it with the report: at 8% tax, 400% becomes 432%.

On the third question, find out whether the client wants that money spent or is only allowing it. A higher budget with the target unchanged may not get fully spent, and that's fine. Forcing spend pushes money toward the placements that convert least.

We don't diagnose from the email alone. The client's answers go into the math, and the account data settles the question.

Step 3: Check the account against the question

Pull the enabled campaigns for the last 30 days: cost, conversion value, ROAS, target, budget use, and Search impression share lost (budget). Two campaigns sell tents in the video:

• the Search campaign returns 608% against a 400% target, spends 97% of its budget, and loses 38% impression share to budget;
• the Performance Max campaign returns 522% against a 500% target and spends only 69% of its budget.

Performance Max doesn't give a usable impression share lost to budget (the API returns zero), so budget use is the signal there. The table already shows which campaign is short on money and which one has money it isn't spending.

Step 4: Work out break-even ROAS at the sale margin

A discount changes the math twice. It cuts the margin, which raises break-even, and it cuts the value of every order, so the same traffic shows a lower ROAS.

• Break-even ROAS = 1 ÷ margin. A 40% margin (0.40) gives 2.5, or 250%. A 25% margin gives 4, or 400%.
• The same orders at sale prices are worth 25% less, so estimated sale ROAS = current ROAS × 0.75. That holds when the campaign sells only sale items. With mixed sales, apply it to the sale items' share.

Check the client's margin figure with the math. Margin after a discount is (margin − discount) ÷ (1 − discount): a 25% discount on products with a 40% margin leaves only 20%. A higher number from the client usually means the supplier funds part of the discount or the stock was bought cheaper. Ask, because at 20% break-even jumps to 500% and the whole plan changes. In the video we use 25%, the figure the client gave.

Performance Max drops from 522% to about 392%, just under break-even. Search drops from 608% to about 456% and stays above it. The × 0.75 estimate assumes bids don't move. Smart Bidding holds the target against post-discount values, though, so with a 500% target it bids down and gives up some volume to get back toward 500%. So the question "why not lower the target and sell more?" has a short answer: anything below 400% ROAS loses money, and Performance Max is already estimated below break-even at its current target, so there's no room to lower it. The estimate is also conservative, since it ignores the higher conversion rate a sale usually brings.

In our target ROAS guide we say a sale is a time to lower or remove the target, and that a campaign that won't spend its budget usually has a target that's too high. Both rules have a floor: break-even at the sale margin. Lower the target no further than break-even, and preferably keep some headroom, because target ROAS is an average and individual sales will land below it. Removing the target only makes sense when the sale margin leaves plenty of room above break-even. Performance Max spends only 69% of its budget here, and the target stays anyway, because its estimated sale ROAS is already below break-even.

Step 5: Add budget only where it's missing

A budget increase helps only when the campaign spends its full budget and still loses impression share to budget. The Search campaign meets both conditions, so it goes from $80 to $130 a day for the ten days of the sale. That's $500 extra against the $1,500 cap the client gave. Performance Max stays as it is, because it doesn't spend what it already has.

Outside a sale, we raise budgets 10-20% every 50 or so conversions. For a short sale, a 62.5% jump is deliberate, because it comes with a rollback date. Google can spend up to twice the daily budget on a single day, and it caps monthly charges at 30.4 times the average daily budget, so heavy days even out on the invoice. $500 is a realistic plan, and the room left under the client's cap stays available for a further increase. Sale traffic often peaks at launch and again right before the end, so on a longer sale you can weight the budget toward those days.

Step 6: Prices, product feed, and sale messaging

The Search campaign sends people to the site, so we raise its budget once the sale prices are live on the site. Performance Max and Shopping ads show the price from the product feed, so there the feed is the trigger. In the video both happen Friday morning. The discounted price goes in sale_price, the sale window in sale_price_effective_date, and price stays the regular price. A feed price that doesn't match the site or the checkout can get the product disapproved in Merchant Center, right on launch day.

You can't force the sale annotation in Google Shopping, and you can't force a particular sitelink to show. Tell the client up front, or they'll assume someone dropped the ball.

We announce the sale on the Search campaign with a sitelink asset pointing to the sale page, with text the store approves and an end date set on the asset so it stops by itself. Google Ads also has promotion assets, with the discount and dates built in, and Merchant Center has its own promotions for Shopping. Both take longer to review than a sitelink, so set them up early. A sitelink added at the campaign level overrides account-level sitelinks in that campaign, so check what the campaign loses before you add one. Don't swap assets in a running Performance Max campaign for the sale, because the system starts re-testing combinations and performance can wobble mid-sale. Have anything that needs Google's review ready two or three days before launch, because review can drag in busy seasons.

Step 7: Preview and approval before every change

Every change starts as a preview: the campaign, the value before and after, and the percentage change. The preview in the video shows +62% (62.5% exactly) and warns that Smart Bidding needs time to adjust to a change that size. We plan for one to two weeks to be safe, and on a ten-day sale that matters. That's why the increase goes to the campaign with comfortable headroom above break-even: it can absorb a few unsettled days.

The change waits for a person to approve it and for the sale to start. The client should know who makes changes to the account, so nobody on their side applies Google's recommendations that touch the same settings in the meantime. Auto-apply recommendations stay off.

Step 8: Stop condition and rollback date

The Search campaign has a 400% target, exactly the sale break-even. Average ROAS of about 456% leaves headroom, but the extra $50 a day buys pricier clicks. A 400% target is an average, so it lets the campaign keep buying clicks right down to break-even, and some below it. So during the sale we check ROAS cumulatively from launch every day and compare it with the 400% break-even. The last three days only serve as an early warning: at two or three conversions a day that's too little data to act on, and recent days read low because some conversions report late. If cumulative ROAS closes in on break-even, we raise the campaign's target or pull the budget back. That stop condition goes into the plan before launch, along with who checks it.

The rollback date goes into the same plan: the Monday after the sale, the budget goes back to $80 and the sale sitelink stops. With the end agreed in advance, nobody has to remember it on a day when the store has moved on to something else.

Sale data is an outlier. Don't set a target from a window that includes the sale, and after the rollback, wait for about 50 new conversions before judging the campaign. A long season is different: there the budget comes down gradually, 10 to 15% a week. After a short increase for a sale, the budget goes back in one step.

Step 9: What the reply says

Keep the reply to a few lines, in plain language, the way the client talks about their store. Impression share and bid strategy names stay in the analysis. The email carries:

• the plan with numbers: what changes, for how long, and what it costs against the client's cap,
• one sentence on what you're not changing, if the client might have expected it,
• requests to the client, numbered so they can answer point by point,
• what happens if the client doesn't reply in time,
• the rollback date, written as an actual date.

Here's the reply from the video:

"Hi, thanks for the heads-up on the sale. Here is our plan for the ten days. We raise the Search budget for tents and sleeping bags from $80 to $130 a day. That adds $500 in total, well inside your $1,500 ceiling. The target ROAS stays at 500%: at the sale margin you break even at 400%, so a lower target would sell at a loss. We need two things from you: 1. Please confirm that stock of your best-selling tents can handle more orders. 2. Please approve the text for a sale sitelink, which we will send separately. Everything goes back to normal the Monday after the sale."

That draft needs four edits. The 500% sentence is about the Performance Max campaign, and a client who opens the Google Ads UI will see a 400% target on the Search campaign. "A lower target" should read "a target below 400%", since the loss only starts below break-even. It needs one line on what happens without an answer, for example: "If we don't hear back about stock by Thursday, the budget stays as it is." And "the Monday after the sale" should be a date.

Keep certainties and estimates apart. "Sale ROAS of about 456%" stays in the analysis, and if it goes into the email, label it as an estimate.

Step 10: Choose the format

In the video the reply comes out as plain text. This step and the next ones cover how to get it to the client. The format depends on how many numbers the reply carries and what the client will do with it:

• plain text works when there's no table,
• HTML in the email body fits one or two tables, and still works if the client forwards the email,
• a reply page suits a longer analysis with several tables and a chart, when the numbers will keep changing,
• a PDF still fits a formal report the client files or forwards to leadership.

Columns lined up with spaces only look right where they were typed. Gmail and Outlook use proportional fonts, so the table scatters. Whatever the format, the reply can land in the inbox directly as a draft in the client's thread (step 13).

Step 11: An HTML table to paste into the email

The AI assistant builds the reply as a single HTML file. We open it in a browser, click the "Copy reply" button, and paste into a new message. Here's the instruction we use:

"Build the client reply as a single HTML file. Put tables in <table> with inline styles on each tag, with no <style> block and no classes. Keep it under 600 pixels wide, right-align the numbers, include the currency with every amount, and use a light background with dark text and no images. Above the reply, add a box with instructions and a "Copy reply" button that copies only the reply, without the box. Take numbers only from the data pulled in this conversation."

The button copies the reply alone, so the instructions don't end up in the client's inbox, which is easy to miss when you select the whole page. The last sentence keeps a number from memory out of the reply. Styles have to be inline because a <style> block only partly survives pasting, and some email clients drop it. The light background matters because dark-mode apps can invert colors. Classic Outlook for Windows renders email with Word's engine: tables and inline styles survive, while modern CSS layout like flexbox and grid breaks. Gmail clips messages larger than about 102 KB and hides the rest behind a link, so a long reply belongs on a page.

Step 12: A reply page

A longer analysis goes on a page, and the client gets a link in the email instead of an attachment. You can update the page at the same address when new numbers come in, so put the date the data was pulled at the top, with a short change log underneath. Without it, the client sees different numbers than last time and has no idea why.

Privacy takes more than a hard-to-guess address. Links in email get forwarded, and email security scanners open them, so a page with sensitive data needs a login. Whatever the setup, set it to noindex, keep the client's name out of the address and the page title, and show only the numbers the decision needs. If the page sits with a third-party host, check that your agreement with the client allows it. Take the page down once the matter is closed.

Step 13: A draft in the client's thread

This is the shortest route: an AI assistant with mailbox access saves the reply as a draft in the client's thread, HTML table included, and a person reads it, edits it, and sends it. The draft only lands in the right thread if it keeps the same subject and references the client's message. Otherwise it starts a new thread, so check that along with the recipients before sending.

Claude doesn't need a separate integration for this. Its built-in Gmail connector turns on under Settings, in the Connectors section, or with /mcp in Claude Code. The assistant finds the thread by sender or subject, reads the whole conversation, and saves the draft as a reply to the client's latest message, with the same recipients and subject. A message link copied from the browser usually isn't enough for it, so give it the sender and the subject. The connector sees attachment names but can't open the files, so save a client's spreadsheet to your computer first. The draft lands in the mailbox the connector is signed into. If the client wrote to a different address, check the sender and the signature before you hit send.

"Find the latest email from [client] about [topic] in Gmail, read the whole thread, and save the reply as a draft in that thread. Don't send anything."

In Gmail, the permission to create drafts also covers sending. Keeping the assistant to drafts has to come from the tool that gives it mailbox access, one that can only save drafts. Sending stays a human decision, just like a change to the account: preview first, then a yes. Claude's built-in Gmail connector can send email too, so the instruction spells it out: save a draft and send nothing.

Step 14: Check before you send

• The numbers in the email match the change preview ($80 → $130, +62% in the preview).
• Currency is stated, and it's clear whether amounts include tax.
• "The Monday after the sale" is replaced with an actual date.
• No jargon: impression share and bid strategy names stay in the analysis.
• The email is consistent with your last email to this client, and any change of position is named outright.
• Estimates are labeled as estimates, and requests are numbered.
• You sent it to yourself first and opened it on a phone, in dark mode too. The reply page link opens in a private window.
• It's the right thread and the right recipients.

Rules at a glance

• Every sentence in a client email hides a question. List them before you look at the data.
• Ask three questions at most, and start with margin. Confirm what margin covers and whether conversion value includes tax. Turn "budget is not a problem" into a number.
• Break-even ROAS = 1 ÷ margin. A discount raises break-even and lowers ROAS on the same orders. Check the client's sale margin with the math.
• During a sale, lower the target no further than break-even at the sale margin, and preferably keep headroom. Budget first, target second.
• Give more budget only to a campaign that spends its full budget and loses impression share to budget. For Performance Max, look at budget use.
• Start Search once sale prices are on the site, Performance Max once they're in the product feed. sale_price with sale_price_effective_date, price untouched.
• Give sale messaging an end date. Leave a running Performance Max campaign's assets alone during the sale.
• For every change: preview, approval, and a stop condition and rollback date set before launch.
• Keep the reply to a few lines: the plan with numbers, one sentence on what stays, numbered requests, what happens without an answer, the rollback date.
• Send tables in email only as HTML with inline styles. Put longer analysis on a page with a date and a change log. A person always hits send.