Google has published in the Google Ads Help Centre a page titled About Review summaries, spotted on 24 October 2026 by PPC News Feed. It documents customer review summaries written by Google's AI and inserted into Search ads, in English, in the United States, the United Kingdom, Australia and India, without the advertiser being able to change a single line. A company's online reputation thus enters the advertising space it pays for, and Google holds the pen. The format does not exist yet in Luxembourg; reviews, however, already feed Google's AI in the country's results.
The facts
- The help page, undated, describes summaries that Google's AI automatically aggregates from reviews found on the web, without naming the source platforms, and displays inside the ad as a short comment. It states twice that these summaries are positive only, invokes Google's responsible AI practices and a requirement of grounding in the original reviews. Scope: Search campaigns, all positions on Google.com, all devices, in English on four markets; all verticals except sensitive categories, pharmaceuticals among them, the other exclusions being unlisted.
- Two rules weigh more than the others. Google shows the summary only when it predicts a performance gain for the ad, without saying on which metric. The text cannot be edited or customised, since it represents, according to Google, the voice of users. The page lists four expected benefits, from customer trust to click quality, with no figure, no test and no client case. PPC Land (25 October) notes that it mentions no opt-out, no review threshold, no recency rule, and that the Help Centre's AI assistant suggests the question how do I enable AI review summaries, which the page never answers.
- The format took shape over nearly a year of tests, according to PPC Land's timeline: review excerpts in Search ads spotted on 3 December 2025 by Anthony Higman, mostly in e-commerce; generated summaries under ad descriptions reported on 3 July 2026 by Darcy Burk, with a warning about possible mistakes; a What Customers Love block observed by Search Engine Watch on 31 August 2026, labelled AI-generated from the store ratings and reviews, then a variant generated from user reviews across the web; the same block seen on 5 September on a Semrush ad. On the day the help page was reported, PPC News Feed also reported a test of a scrolling display of ratings and reviews in ads (Search Engine Roundtable, 26 October).
The help page describes a positive, non-editable text, shown when Google sees a gain
The table below sets out, point by point, what Google's page establishes and what it leaves open. The grey areas count as much as the rules: an advertiser running on these markets knows neither where the summarised reviews come from, nor when the summary shows, nor what it produces.
| Point | What Google says | What the page does not say |
|---|---|---|
| Content | Positive-only summaries, generated from reviews aggregated across the web, grounded in the original reviews | Source platforms, number of reviews required, recency, handling of a namesake |
| Display | Search campaigns, all positions on Google.com, all devices, when Google predicts a performance gain | The performance metric used, the display frequency |
| Markets | In English in the United States, the United Kingdom, Australia and India | Any date, any extension to other countries or languages |
| Verticals | All, except sensitive categories, pharmaceuticals among them | The list of excluded sensitive categories |
| Control | Text that cannot be edited or customised, as it represents the voice of users | Any opt-out; the link to account-level automated assets does not say whether it applies |
| Measurement | Four expected benefits: trust, click quality, performance, brand image | Any figure, any test, any report showing when the summary was displayed |
A format that falls outside the text guidelines framework
Google already writes part of the ads. Since 12 February 2024, its generative AI has produced automated headlines and descriptions for English-speaking advertisers in the United States and the United Kingdom; those assets were enabled campaign by campaign and the advertiser could remove the ones that did not suit. Text guidelines, introduced on 10 September 2025 and extended to all languages on 26 February 2026, allow up to 25 term exclusions and 40 messaging restrictions per Performance Max or AI Max campaign, as we wrote about AI Max's promises. Review summaries fall outside this framework: the help page links to account-level automated asset settings without saying they apply to this format, and mentions neither text guidelines nor term exclusions.
The advertiser pays for the placement; the text, presented as the voice of its customers, is compiled by Google. No report is announced to know when a summary was shown, or with what effect. We have been running campaigns long enough to know the fate of a format that can be neither switched off nor measured: it ends up explaining a click-through rate variation in a monthly report, without anyone being able to prove it. The same reflex applies to the promotions Google Ads picks up from the website since 12 October: look at the ad as it is displayed, as well as the console.
Positive-only summaries: what studies measure about the tone of AI summaries
A flattering summary, displayed at no extra cost in one's ad, looks like a gift. The available research invites a second look. A study from the University of California San Diego (Alessa et al., presented in December 2025 at the IJCNLP-AACL conference, reported by Live Science on 14 March 2026) had 70 participants read either original reviews of electronics products or their summary by a language model: the former said they would buy in 52% of cases, the latter in 84%. The six models tested changed the tone of real reviews in 26.5% of cases. The intention is declared, in an experimental setting; no sale was measured.
At Virginia Tech, Joanna Quan and Wenqi Shen presented on 19 March 2026 a study titled Soft Deception by Design, filed with the Federal Trade Commission, on 905 hotels in 21 US cities and 96,630 Tripadvisor reviews published between November 2024 and October 2025. On a scale from -1 to +1, AI summaries show an average tone of 0.94, against 0.71 for the reviews they summarise. The gap peaks for hotels rated between 2.5 and 3 out of 5 and for those with few reviews: the weaker the material, the more the summary embellishes it. The bias also works the other way: in late 2023, Amazon sellers complained about summaries that amplified a minority of negative reviews, according to Bloomberg, relayed by PPC Land. The selection rules remain opaque, whichever way the filter leans.
For the advertiser, the risk lies in the gap. If the ad promises an experience the customer does not find, the disappointment will show in the next reviews, the ones the AI will summarise next time. This loop has not been observed; it follows from the mechanism. Google says it blocked or removed more than 292 million reviews that broke its rules in 2025, and 13 million fake business profiles. A summary is only worth the reviews it reads.
In Luxembourg, the format does not exist, but reviews already feed Google's AI
To date, review summaries in ads are available neither in Luxembourg nor in the European Union, and Google has announced no extension. Part of Luxembourg's advertisers are nevertheless concerned right now: those running Search campaigns in English towards the United Kingdom, the United States, Australia or India. Companies in finance, B2B services, export e-commerce, tourism or international recruitment are the first that need to look at their ads. For the others, the subject is one of anticipation: Google Ads formats often start with English-speaking markets, and nothing allows a European arrival to be dated. Luxembourg's framework for reviews already exists. The law of 30 November 2022, which transposed the European Omnibus directive into the Consumer Code (bill 7904, voted on 23 November 2022), requires sites that publish reviews to guarantee that those reviews come only from people who bought or used the product, according to the government's press release. How a summary written by Google, from sources it does not name, would fit with that obligation remains a question for a lawyer, the day the format arrives.
Everywhere, reviews published on Google Business Profile and on sector platforms already shape the way Google's AI presents a company. On 27 October, we asked sixteen queries from Luxembourg for our article on AI Overviews: all sixteen showed an AI summary in an everyday browser, and the two review queries about insurers were summarised from a third-party comparison site and Trustpilot, without the insurers' websites. Advertising summaries extend this logic into the paid space: in Luxembourg, the raw material is already read by the AI, in the free results.
Five actions to put in place now
The actions below are our own practice; they apply to an advertiser running in English on the four markets as much as to a company preparing itself.
- Treat your reviews as an advertising asset. Take stock of the volume, average rating, recency and recurring themes in your reviews, on Google and on your sector's platforms. That is the raw material of the summary.
- Structure the collection and reply. Ask for a review after each purchase or service, without incentives contrary to the platforms' rules. Reply to positive reviews as well as negative ones: the corpus the AIs read becomes more complete, and your future customers read it too.
- Look at your ads if you run in English on these four markets. Take regular screenshots of your ads as they are displayed, check the account-level automated asset settings and ask your Google contact whether an opt-out is possible. The documentation mentions none.
- Get customer service, local SEO and SEA working together. A shared dashboard, reputation and performance, prevents each team from discovering the other's problem too late.
- Appoint someone to watch what AI says about your brand. Advertising included. Someone in your company must know what the AI writes in your name, in result summaries as in ads.
A negative review left unanswered, an untended Google profile, a forgotten review platform: all of that existed before this format. What changes is that Google can now turn it into the text of an ad the company pays for.
AIxH's view
Platforms write a growing share of ads, and this one has a particularity: it feeds on what customers think of a company. Running a campaign therefore extends beyond the advertising console. This is where the work of our Google Ads team in Luxembourg meets the GSO audit: one looks at what the ad actually displays, the other at what Google and AI assistants say about the brand, and from which sources. If you run in English towards the United Kingdom or the United States, looking at your ads as they are displayed is the first check to make; if you advertise in Luxembourg, your Google reviews are already read by the AI, and that is where we start.
