On this page13
- What Was Reported
- Why a Summary Changes the Question
- What the Report Does Not Say
- The Label That Went Missing
- How to Check a Citation Card by Hand
- How to Fix AI Hallucinations About My Brand in a Summary
- A Two-Week Test Plan
- Mistakes Agencies Will Make
- What AI Visibility Tracking Must Read
- A Report Line for Citation Cards
- Questions Clients Will Ask
- What to Tell Clients
- Frequently Asked Questions
Google appears to be testing AI-written summaries inside the citation cards beside AI Overviews. Search Engine Roundtable reported it on October 23, 2026. The card may show Google’s summary of a page, not the page’s own words. AI visibility tracking now has to ask what the card says about the client, not only whether it links.
What Was Reported
Barry Schwartz of Search Engine Roundtable reported the change on October 23, 2026. Brodie Clark spotted it and posted on X. Schwartz confirmed it on some of his own test queries, but not all. The article says Google appears to be “testing or rolling out” the change, and it quotes no statement from Google.
| Question | What the report says |
|---|---|
| What changed? | Citation cards beside AI Overviews can show an AI-generated summary |
| What did cards show before? | Original page text, per the report’s description of the change |
| Who spotted it? | Brodie Clark, on X |
| Who confirmed it? | Barry Schwartz, on some of his queries and not others |
| How is the summary chosen? | Clark: “it is being summarised based on the information that Google thinks is most relevant to the query” |
| Has Google commented? | The article quotes no statement from Google |
Schwartz added a dry remark: “We got the AI, generating more AI, for more AI...” The tone is light. The point for an agency is not.
Why a Summary Changes the Question
Until now, a citation card was a pointer. It named a source and showed some of that source’s text. A reader who clicked got the page as written. A summary adds a layer. Google decides which parts of the page matter for this query and writes a new sentence.
Everything in this section is an inference, not a finding in the report. It follows from what a summary is. A summary can leave out the detail a client wants shown. It can also restate a fact in a slightly different way.
| Feature | Card with page text | Card with a Google summary |
|---|---|---|
| Who wrote the words | The client | Google’s system |
| What decides the content | The page | The page and the query |
| Chance of changed meaning | Low | Higher, because words are rewritten |
| What the agency checks | That the page is correct | That the page and the card both say the right thing |
The last row is the practical change. A correct page no longer guarantees a correct card.
What the Report Does Not Say
| Question | In the report? | What it means for a client |
|---|---|---|
| How many queries show the summaries? | Not stated. Some of Schwartz’s did, some did not | Test, do not assume |
| Is it a test or a full rollout? | “Testing or rolling out” | Treat it as unstable |
| Does the summary ever change a fact? | Not stated | Unknown, so check |
| Does it affect clicks? | Not stated | Watch Search Console |
| Which countries and languages? | Not stated | Test each market |
Do not tell a client that summaries misstate pages. The report does not show that. Tell them that the card is now a place to check.
The Label That Went Missing
A second, smaller story from the same week. Schwartz reported on October 20 that some AI Overviews lacked the “AI Overview” label. Glenn Gabe spotted it on desktop and wrote: “No AIO label? -> Very interesting to see AIOs for shows or movies and Google not provide the AIO label.” The label still showed on mobile for the same feature.
| Detail | What the report says |
|---|---|
| Where it showed | Desktop results lacked the label. Mobile still showed it |
| Which results | Mostly AI Overviews at the top of results that resemble knowledge panels |
| What happened next | On October 22, the label was restored |
| Cause | Schwartz’s guess is a bug. He wrote “I guess it was a bug” |
| Google’s comment | None quoted. He reached out and reported no response |
The lesson is small but real. A rule that says “count only labelled AI Overviews” would have missed some for two days. Track what appears, not what the page labels.
How to Check a Citation Card by Hand
- Pick ten buyer questions per client. Run them in Google and open each AI Overview.
- Open the citation card for the client page. Copy the text exactly.
- Place the card text beside the matching text on the page.
- Mark each claim as faithful, changed or missing.
- Save the card text, the query, the date and the device.
Step five matters because the report shows the behaviour was not on every query. A saved record shows when a card said what. SKYA explains why dated records matter in its note on AI hallucination monitoring over time.
How to Fix AI Hallucinations About My Brand in a Summary
If a card misstates the client, the fix still starts with the source. An agency cannot edit Google’s summary. It can improve the page the summary draws on. What follows is guidance, not a finding from the report.
| Problem in the card | Likely cause on the page | Fix |
|---|---|---|
| A key fact is missing | The fact sits low on the page | Move it to the first lines under a clear heading |
| A fact is stated loosely | The page words it loosely | State it in one plain sentence with the exact figure |
| An old fact appears | An older page is still live | Update or remove the old page and keep dates visible |
| A rival’s fact is attributed to the client | Pages blur two products | Separate the pages and name each product clearly |
To fix AI hallucinations about my brand, a client needs the wrong claim, the date and the cited page in one place. That is a stored reading. SKYA covers the audit in its page on AI hallucination.
A Two-Week Test Plan
One sighting proves little. A short, structured test tells you whether the change matters for your clients.
| When | What to do | What to record |
|---|---|---|
| Day 1 | Run ten buyer questions per client and open each citation card | Card text, page text and the device used |
| Day 3 | Repeat the same questions | Any change in card text |
| Day 7 | Repeat again and add a second device | Differences between desktop and mobile |
| Day 14 | Compare all readings | How often cards carried a summary, and how faithful it was |
Keep the wording of each question fixed. If you change the words, you change the test. At the end of the two weeks, you can tell a client how often it happened, on which questions and whether the summaries held up.
Mistakes Agencies Will Make
| Mistake | Result | Better move |
|---|---|---|
| Checking only the page, not the card | A wrong card goes unseen | Compare both |
| Assuming every query shows summaries | Tests that do not match buyers | Test the client’s real questions |
| Filtering by the AI Overview label | Missed answers during label gaps | Record every AI Overview that appears |
| Reporting one sighting as a trend | A false story for the client | Repeat the checks and keep dates |
What AI Visibility Tracking Must Read
Good AI visibility tracking reads the answer, not only the link. A summary in a card is part of the answer. Ask any vendor the five questions below.
| Requirement | Why it matters | How SKYA handles it |
|---|---|---|
| Keeps the answer text and date | You need to compare cards over time | A stored reading behind every number, with an evidence panel |
| Reads Google AI Overview and AI Mode | Citations appear in both | Nine assistants every day: ChatGPT, Gemini, Perplexity, Claude, Google AI Overview, Google AI Mode, DeepSeek, Grok and Copilot |
| Checks claims | A card can state a wrong fact | A hallucination audit on every paid plan |
| Admits what it cannot read | A guessed score hides gaps | No answers read means no figure. A run that measured nothing gives its credit back |
| Fits agency work | Many clients, one report | Multi-client workspaces and white-label PDF reports |
One more question is worth asking of any vendor, SKYA included. Does the stored reading include the text shown in the citation card, or only the main answer? The report says the cards are changing, so the answer matters. SKYA publishes how every number is calculated.
A Report Line for Citation Cards
| Section | Source | What the client learns |
|---|---|---|
| Mentions and citations | Daily answer readings | Whether the client is named and cited |
| Card text check | Manual comparison of card and page | Whether the summary is faithful |
| Accuracy | Claim check on stored answers | Which statements were wrong |
| Fixes made | Agency notes | Which pages changed and when |
| Open questions | The report’s gaps | What we cannot yet say |
Questions Clients Will Ask
| Question | Plain answer |
|---|---|
| Is Google rewriting our content? | It appears to summarise pages in some cards. The report says it is a test, and not on every query. |
| Will it hurt our traffic? | The report does not say. We watch Search Console and report what we see. |
| Can you control the summary? | Not directly. We make the page clear so the summary has solid text to draw on. |
| How will we know if it gets a fact wrong? | We compare the card with the page and keep dated records. |
What to Tell Clients
| Say | Avoid |
|---|---|
| Google appears to be testing AI summaries in the cards beside AI Overviews. We are checking how yours read. | Google now rewrites your pages. |
| It does not show up on every query. We test the questions your buyers ask. | Every AI Overview works this way now. |
| If a card misstates a fact, we fix the page it draws on and keep the dated record. | We can edit what Google shows. |
| One report of a test is not a trend. We will repeat the checks and tell you what we see. | This will cut your clicks. |
Frequently Asked Questions
Is Google showing AI summaries in AI Overview citation cards?
It appears so. Search Engine Roundtable reported on October 23, 2026 that Google seems to be testing or rolling out AI-generated summaries in the citation cards beside AI Overviews. The article quotes no statement from Google.
Do the summaries appear for every search?
No. Barry Schwartz saw the behaviour on some of his test queries but not all. Test the questions your own buyers ask.
Why did some AI Overviews lose their label?
Search Engine Roundtable reported on October 20 that some desktop AI Overviews lacked the label, and that Google restored it on October 22. The author guessed it was a bug. Google gave no comment in the article.
How do I fix AI hallucinations about my brand in a card?
Compare the card text with the page, find the cause on the page and fix the source. State key facts early and plainly, remove old pages and keep a dated record of each reading.
What is AI visibility tracking?
It is the daily reading of what AI assistants say about a client, with dates and a claim check. SKYA reads nine assistants every day and stores the answer behind every number.
Written by
SKYA Editorial Team
The SKYA Editorial Team publishes field-tested research on Answer Engine Optimization, Generative Engine Optimization and AI Visibility. Every article is reviewed by senior practitioners from the SKYA product and research groups before publication.
Related posts
Keep exploring the SKYA Journal.
