Google Now Writes AI Review Summaries Into Search Ads: What It Means for Your On-Site Summaries

Around October 24, 2026, search marketers noticed a new Google Ads help page titled "About Review summaries." The page says Google’s AI reads online reviews of a business, writes a short summary, and places it inside that business’s Search ad. Two details stand out. According to PPC Land's reading of the page, the summaries are positive only, and advertisers can't edit a single word.

If you sell on Magento 2, that sentence should make you pause for a different reason than it does for advertisers. A shopper may now see Google’s version of your reputation first, written by a machine, in your paid ad. Then they land on your product page, where you may be showing your own AI-written review summary. The two need to tell the same story, and yours needs to be the more honest one.

This guide covers what Google has and hasn’t said, why a positive-only summary raises the bar for everyone else, and how to audit your on-site summaries. The examples use MageDelight’s AI Review Summary and Customer Sentiment Analysis extension. Some of what follows is still unknown, and those parts are flagged as we go.

What Google Has Actually Said

The details below come from PPC Land’s detailed report, which quotes Google’s help page. The writer could not load the page directly, so read the page itself before you act on anything here.

It didn’t come from nowhere. PPC News Feed spotted review excerpts in Search ads on December 3, 2025, AI summaries under ad descriptions on July 3, 2026, and tests labeled "What people are saying" and "What customers love" in August and September. The help page is the first time Google has described the feature in its own words.

Why This Matters Even If You Don't Advertise in Those Four Markets

There are four reasons, and only the first one is about ads.

The ad now makes a claim you didn’t write. If Google tells customers you ship fast and your delivery times are uneven, shoppers expect your page to match those claims. The difference between what the ad promised and what the page actually delivers is where trust disappears.

Summaries change decisions, and models drift. A University of California San Diego team had large language models summarize product reviews, then asked 70 people about buying. Those who read the machine summaries said they would buy in 84% of cases, against 52% for those who read the originals. Be careful with that number: the researchers chose examples where the model's perspective changed a lot. They said the test was not very important. So the paper discovered that the models changed the feeling of the original text in about a quarter of the summaries. They also showed a preference for what came in the text over the criticism that came later. Your on-site summary tool is a language model too. It can drift the same way.

Google reads what’s out there. Whatever feeds its summaries comes from across the web. You can’t control the text, but you can control what customers have to say about you, where, and how often you reply.

The rules about reviews apply to you, not to Google. The FTC’s rule on fake reviews and suppression covers how a business displays reviews. If your page implies that what you show represents most of the reviews you received, and you filtered by rating or sentiment, that's what section 465.7(b) addresses, as we explained in our FTC review rule audit. Google decides the positive-only ad summary. A positive-only summary on your own page is yours.

What a Good On-Site Summary Looks Like Now

Think of it as the honest counterpart to Google’s highlight reel. It should show both sides in proportion, use plain language, say it was written by AI, and show when it was last updated. It should never contain a fact that no review contains. And it needs a human spot-check, since nobody is auditing it for you.

The extension is built around that idea. Per the user manual, last modified in August 2025, each product page can show an AI Analysis section under the Reviews tab with: the average rating and review count, a sentiment badge, a short summary, a sentiment distribution bar (positive, neutral, negative), a "What Customers Loved" list and a "What Could Be Better" list with mention counts, a last-updated date and an optional Powered by AI note. The "What Could Be Better" list is what separates it from a highlight reel, and you should turn it on.

The Extension Settings That Decide Whether It's Honest

Two caveats. The manual lists Google Gemini and OpenAI as providers, while the product page also mentions Claude, so check which your version supports. The manual also doesn’t describe a human approval step before a summary goes live, or a way to edit a summary by hand. Instead, the admin offers Reprocess and Delete per product. That makes your own sampling more important. The FAQ page is a good place to ask MageDelight support about both.

A Seven-Step Audit

Step 1: See What Google Might Say About You

Search for your brand and your top ten products in an incognito window. If possible, use a VPN or ask a colleague in the US, UK, Australia, or India to do the search. Look carefully for any review text that appears under the paid results. If you are running Search ads, take screenshots every week during the holiday period. Then list the sources where Google might be getting reviews: Google Business Profile, shopping and marketplace listings, Trustpilot or similar review sites, social media posts, and your own website pages. For each source, list the three most common praises and the three most common complaints.

Step 2: Compare Those Themes With Your Page

Put the themes next to what your in-person summary says for the same products. Any theme that shows up outside but never on your page is a flag. If customers complain about sizing and your summary doesn't mention sizing, someone will notice.

Step 3: Fix the Settings

Work through the table above on staging. Turn on both lists and the distribution chart. Raise the minimum review count. Clear the older-than filter. Set the Powered By text. Then reprocess a handful of products and read the output with the original reviews open next to it.

Step 4: Write the Prompt Like a Style Guide

The template accepts the placeholders {review_title} {review_text}. Rating_value}. A short, firm prompt works better than a clever one. Something along these lines:

Summarize the customer reviews below in plain, neutral language.

Keep the balance of opinion: if most reviews are positive and some are not,

say so. Always mention the most repeated complaint, even if it is minor.

Use only facts from the reviews. Do not add features, claims

or advice. Do not use marketing words such as "perfect" or "amazing".

Say how many reviewers mentioned each point when you can.

Reviews: {review_title} {review_text} {rating_value}

Treat this as a starting point. Test it. Don't expect prompts to control behavior. The UCSD team tried 18 ways to fix the problem, but none of them removed bias from every model and every task.

Step 5: Audit the Output on a Schedule

 Select 20 products: some of your sellers, some of your lowest-rated items, a handful chosen at random, and a few that have more than 50 reviews. For each product, read a sample of the reviews. Evaluate the summary on four aspects. Is the overall mood correct? Does the summary contain the complaint? Are all claims from the reviews? Is the summary brief enough to read? Aim for at least 18 of 20 to pass. If the result is below that threshold, adjust the prompt or the word limit, reprocess the summaries, and test again. Repeat monthly, and whenever you change the prompt or the provider. This number is a suggestion, so set your own bar.

Step 6: Use the Negative Signals

The extension’s Top Negative Products and Most Mentioned Cons reports tell you where customers are unhappy. Fix the cause where you can: a size chart, a photo, better packaging, space, a delivery promise you can keep. Reply to reviews politely. Better products and better replies change what Google reads, though nobody outside Google can promise how.

Step 7: Line Up the Ad and the Landing Page

• Show the star rating, show the review count and show a link to the reviews above the fold on the landing pages for your main ad groups.

• Read your ad copy next to your landing page. Don’t promise anything the page doesn’t back up.

• Track clicks on the Reviews tab and bounce rate by landing page in GA4 with GTM, and tag campaigns consistently with the Campaign Performance Tracker.

• Ask your Google representative or Google Ads support whether any control exists for review summaries in your account, since the help page describes none.

Keep Your Review Supply Healthy

Summaries are only as good as the reviews behind them. Short one-liners give a model little to work with. Structured reviews with pros and cons, photos, and verified buyer badges give it more. Advanced Review Ratings & Reminders supports that kind of review, along with reminder emails. Keep reminders neutral, send them to every buyer, and don’t tie rewards to a rating. Our FTC audit covers those rules in detail. If your summaries need to read well in several languages, the AI content workflow shows how to review translations.

Common Mistakes

• Publishing and forgetting. AI summaries drift. Schedule the audit.

• Hiding the negative list because it "looks bad". That is the version of your page that most resembles Google’s highlight reel.

• Letting very few reviews drive a summary. Raise the minimum.

• Skipping new reviews. An older-than filter left on can ignore this month’s complaints.

• Prompting for enthusiasm. Asking the model to sound persuasive biases the output.

• Treating Google’s summary as a problem for the ad team alone. Support, product, and merchandising teams all affect what it says.

Which Setup Fits Your Store

• You advertise in the US, UK, Australia or India: do Steps 1 and 7 first, this week. The summaries may already be running.

• You advertise elsewhere: the feature is documented only for those four markets, but the audit in Steps 2 to 6 is useful anywhere.

• Few reviews per product: hold off on AI summaries until there is enough text, and focus on collecting honest reviews.

• Sensitive categories such as health products: Google says they are excluded from the ad feature, but regulators watch review claims closely. Keep the on-site summary conservative and get legal review.

• Large catalogs: automate the sampling. Pull the lowest-rated 50 and a random 50 each month. A Digital Marketing Audit or MageDelight’s support service can help set it up.

For most stores, the order is: look at what Google may say, fix the settings, write a better prompt, audit monthly, and line up the ads. The AI Review Summary extension handles the summary layer. You still own its honesty.

FAQ

Can I opt out of Google's review summaries in my ads?

The help page, as reported, describes no opt-out and no control over the text. Ask your Google representative or Google Ads support whether any account-level setting applies.

Do the summaries include negative reviews?

No. Google says they are positive only, and they show only when Google predicts they will improve ad performance.

Does this affect my Google Shopping listings or organic results?

The help page covers Search ads only. It says nothing about organic results or Shopping, so don’t assume either way.

Should I remove my on-site AI summary?

Not necessarily. A balanced, labeled, regularly audited summary can help shoppers. A positive-only or unchecked one can hurt trust and may create legal risk. Fix it rather than remove it, unless you can’t audit it.

How many reviews does a product need before a summary makes sense?

The extension’s minimum is configurable, and the manual’s example is 3. For a more reliable summary, set it higher. The right number depends on how varied your reviews are.