BlogFindings
AI Mode citations: five link changes on 15 October, and the only number in the post is a multiple with no base
Google published a Search post on 15 October 2026 naming five changes to how AI Mode and AI Overviews link out to the web, three of them described as tests rather than rollouts. The post gives no number or percentage for the increase in inline links it announces. The only figure in it is that Preferred Sources users have grown more than 7x since the start of the year, stated with no base count and no method. Lantad has measured none of these surfaces and does not claim to.
This post reports that announcement and nothing more about it. Lantad has not measured AI Mode, has not counted citations inside it, and has no visibility data of its own on either surface: the scanner measures what a crawler can fetch and read from a site, which is a different thing and is upstream of all of it. What this blog can add is the part that sits on the reader's own server, where measurement is possible, and that part has been counted recently. Below is what Google said, what its own documentation currently says alongside it, and the three dated corpus measurements that bear on whether a page is in a state to be linked at all. The honest summary is that the announcement changes the shape of the answer page and changes nothing a site owner can check, and the gap between those two is where most AI visibility advice goes wrong.
In short
- Google's Search post of 15 October 2026, bylined Hema Budaraju, VP Product Management, names five changes to how AI Mode and AI Overviews link to the web: more inline links surfaced higher in responses, a tested section of official links at the top of a response, news carousels reaching additional locations and languages, related article links made more prominent with image thumbnails, and a test that makes creator and publisher names on video results clickable through to Search profiles.
- AI Mode citations gained no published quantity in that announcement. Google describes substantial increases in inline links and attaches no number and no percentage to them, and the only figure anywhere in the post is that Preferred Sources users have grown more than 7x since the start of the year, with no base count, no absolute number and no method stated.
- Google's own AI features documentation, which is the page a site owner reaches for after an announcement like this, carried a last updated date of 2025-12-10 UTC when Lantad read it on 17 October 2026, and it describes no control specific to these surfaces beyond the snippet directives, noindex and robots.txt.
- That same documentation states that a page shown as a supporting link must be indexed and eligible to appear in Search with a snippet, and Lantad measured 132 of 1,083 readable corpus home pages sending a snippet directive on 7 October 2026, of which exactly 3 set a value that limited anything at all.
- Search profiles are the destination of the new creator and publisher test, and Lantad measured 2 of 1,093 readable home pages in its 1,419 hostname corpus referencing profile.google.com on 3 October 2026, neither of them in the form Google's own badge guide documents.
Described as launching or rolling out
- More inline links, surfaced higher in responses more often
- News carousels, to additional locations and languages globally
- Related article links, made more prominent with image thumbnails
Described as a test
- A dedicated section of official links at the top of a response
- Clickable creator and publisher names on video results in AI Mode
- Both stated without a date, a market list or a size
What Google announced on 15 October 2026
The source is a single post on Google's own blog, More ways to connect with the web from Search, dated 15 October 2026 and bylined Hema Budaraju, VP Product Management. It has five body sections and each one describes a separate change, so the count of five is the post's own structure rather than an editorial grouping imposed on it here.
The first is inline links. Google says it has made continued, substantial increases to the inline links in its AI experiences and is launching updates that surface them higher in responses more often. The second is a test: a new display that puts a small set of official links in a dedicated section at the top of a response, with context on why each one helps. The third is the news carousel, which Google says is rolling out to additional locations and languages globally, and which highlights Preferred Sources. The fourth makes links to related articles more prominent, with image thumbnails. The fifth is a test, started in the week of the post, that makes creator and publisher names on video results clickable through to their Search profiles in AI Mode.
Two of the five are tests, which in Google's own vocabulary means they may be withdrawn, and neither carries a start date beyond that week, a market list or a share of queries affected. Three are described as launching or rolling out. None of the five is given a size. The phrase doing the work in the first section is substantial increases, which is a direction rather than a quantity, and the post does not say substantial compared to when.
There is exactly one number in the whole piece, and it is not about links. Google says Preferred Sources users have grown by more than 7x since the start of the year. A multiple without a base is not a measurement a reader can use: seven times a small number is a small number, and the post states no starting count, no ending count and no method of counting a user. That feature is not new here either, and this blog read its documentation when the preferred sources button shipped in August 2026.
| Change | Surface | Status in the post | Quantity published |
|---|---|---|---|
| More inline links, surfaced higher | AI Mode and AI Overviews | Launching | None |
| Official links in a top section | AI Mode and AI Overviews | Test | None |
| News carousel in more places | AI Mode and AI Overviews | Rolling out | None |
| Related article links, with thumbnails | AI Mode and AI Overviews | Not stated | None |
| Clickable creator and publisher names | AI Mode, video results | Test, that week | None |
| Preferred Sources adoption | Cited inside the carousel section | Existing feature | More than 7x, no base |
What do the AI Mode citations changes ask a site owner to do?
Nothing, on the evidence of the announcement itself, and that is worth stating plainly rather than filling with advice. The post is addressed to people using Search. It describes a display change. It names no markup, no file, no header and no setting, and it does not link to a documentation page that names one either.
The page a site owner reaches next is Google's AI features and your website documentation, and reading it on 17 October 2026 it carried a last updated date of 2025-12-10 UTC. Ten months is a long time for the reference page to sit still while the surface it describes is announced as changing five ways in one post, and it is the reason the practical answer to this section's question has not moved. That page names the ordinary Search controls and no others: robots.txt for how Googlebot crawls, the nosnippet and data-nosnippet and max-snippet directives, noindex, and Google-Extended, which it ties to training and grounding in other Google systems rather than to these two surfaces.
One sentence on that page does carry weight, because it is the precondition for everything the announcement describes. To be shown as a supporting link, a page has to be indexed and eligible to appear in Search with a snippet. More inline links in an answer is a change to how many eligible pages get surfaced. It is not a change to what makes a page eligible, and a page that an AI crawler cannot fetch, or that returns no readable text when it is fetched, does not become eligible because the answer format gained a thumbnail.
So the useful reading of this announcement for a site owner is narrow and it is defensible: the odds of a link improved by an unstated amount, and the conditions for earning one did not change at all. Those conditions are on your own server, they are checkable today, and the rest of this post is what counting them across a real population of sites has shown. The scanner's own account of how it checks them is on the methodology page, and the crawler that does the fetching identifies itself on the bot page.
-
More inline linksNothing to change A display change to answers. No markup, file or header is named. -
Official links sectionNothing to change A test. Google states no criterion for which pages qualify as official. -
News carousel reachNothing to change Highlights Preferred Sources, which readers set and publishers do not. -
Related article linksNothing to change Thumbnails are chosen by Google from pages already eligible. -
Creator and publisher namesDepends on a Google page Links to a Search profile, which lives on profile.google.com rather than on your site. -
Eligibility to be linked at allYours to check Indexed and snippet eligible is the stated precondition, and it is measurable on your own server.
A supporting link still has to land on a page a crawler could read
The precondition quoted above is the only part of this story that can be counted from outside, and this blog has counted it three times in the last fortnight on the same population: the 1,419 hostnames in the two corpus seed files committed to this repository, requested one home page each as LantadBot with redirects followed.
On 16 October 2026 that run found that 418 of the 1,419 gave a plainly identified crawler nothing to read: 347 delivered no HTML at all, and 71 of the 1,072 that did deliver HTML delivered zero words. On 13 October 2026 a full scan with a real browser attached graded 972 of the 1,419, with 287 refusing a plain fetch, 35 blocking the browser, 35 timing out and 21 failing to resolve in DNS. Those two counts are made differently and on different days, and they agree on the shape of the problem: somewhere between a quarter and a third of this population does not reliably put a readable document in front of an unfamiliar crawler at all.
That matters for an announcement about links because the refusals are not random. On 26 September 2026 a run that changed nothing between requests but the user agent string found 103 of 1,089 sites served an unknown bot and then refused GPTBot, and on 76 of those 103 the site's own robots.txt permitted that crawler at the root. The block was at the edge, not in the file, which means the owner can read their own robots.txt, see permission, and be wrong. Googlebot is not GPTBot and the AI features documentation is explicit that Googlebot crawling is what governs these surfaces, so that figure is not a count of Google refusals. It is evidence that edge rules and file rules diverge often enough that neither can be assumed from the other, which is the whole argument for checking with a request rather than a setting.
None of this is a claim about AI Mode. It is a claim about the state of the web these answers are assembled from, measured on a defined sample with dates attached, and it is the part of the chain a reader controls. The crawlability study is where the population and the method are described in full.
The controls Google documents, and what sites actually send
Of the controls the AI features page names, one applies directly to what an answer may quote, and it is the snippet directive. Google's robots meta tag documentation defines nosnippet, which prevents a snippet being shown, and max-snippet, which caps its length in characters. The AI features page is what connects those two to the surfaces in the announcement, by making snippet eligibility the condition for being shown as a supporting link.
On 7 October 2026 this scanner read the home page of all 1,419 corpus hostnames with no JavaScript executed, and 132 of the 1,083 that answered with HTML sent one of those two directives. That sounds like meaningful adoption until the values are read. 129 of the 132 set max-snippet to minus one, which Google defines as no limit at all, so exactly 3 sites in the whole corpus sent a value that limited anything. The directive is present on about one page in eight and active on three pages in 1,083.
The reading of that is not that sites are getting it wrong. It is that almost nobody is using the one documented lever over these surfaces, in either direction, and most of the files that carry it are carrying a default written by a plugin or a template rather than a decision. An announcement that more supporting links will be shown lands on a population that has, with three exceptions, declined to say anything about how much of its content may be quoted. If a site owner wants a position on that, the directive is where it is expressed, and the what GPTBot sees view is a quick way to check what a stateless fetch of a page actually returns before deciding.
The same pattern showed up on the other Google surface in this announcement. When the Search profile badge guide appeared on 16 September 2026, a capture of the corpus on 3 October 2026 found that 2 of 1,093 readable home pages referenced profile.google.com at all, both inside a sameAs array rather than as the image and link the guide describes. The creator and publisher test announced on 15 October points at exactly those profiles, which is a Google property rather than a page on your site, and it is worth noticing which way that link runs.
- nosnippet or max-snippet in a robots meta tag 132 of 1,083 readable home pages sent one on 7 October 2026, and 129 of those set max-snippet to minus one, which sets no limit.
- A snippet directive that actually limits something 3 of 1,083 on 7 October 2026. One nosnippet and two finite max-snippet values.
- A Search profile reference on the home page 2 of 1,093 readable home pages on 3 October 2026, both inside a sameAs array rather than the documented badge form.
- A readable document for an unfamiliar crawler 418 of 1,419 gave a plainly identified crawler no words on 16 October 2026.
- A Google-specific AI Mode control The AI features documentation, last updated 2025-12-10 UTC, names none beyond the ordinary Search controls.
What can be measured after this, and what cannot
The announcement is a change to a surface nobody outside Google can count directly, and that is the uncomfortable centre of it. There is no public log of which pages AI Mode linked, no sample frame, and no way for a site owner to reproduce a result. What reaches them instead is an aggregate, and this blog read what that aggregate contains: Google's generative AI performance report, read at its own Search Console help documentation on 14 August 2026, offers impressions and no clicks, with no click-through rate, no average position and no query dimension.
Set the announcement against that report and the shape of the problem is clear. Google says links are being shown more prominently and more often. The instrument it gives site owners counts the showing and not the clicking. An increase in prominence is exactly the kind of change that moves clicks without moving impressions, or moves impressions without moving clicks, and the report cannot separate those two outcomes. A reader is therefore being told that something improved and handed a measurement that cannot confirm it. Saying so is not scepticism about the change; it is an accurate description of the evidence available, and it is why this post carries no estimate of the effect.
What can be measured is the other end of the chain, and it is the end that is actually in the reader's control. Whether an unfamiliar crawler receives a document. How many words of prose that document contains before any JavaScript runs, which is the prose parity question and which took 60.7 percent of every point lost across the 972 sites graded on 13 October 2026. Whether the edge and the robots.txt file agree, which 79 of 115 sites got wrong on 13 September 2026. Whether a snippet directive says anything. These are small, dull, checkable facts, and each one is a precondition for the thing the announcement is about.
One external finding is worth keeping next to all of this, because it sets the ceiling on how much of this is ordinary ranking work. Haofei Xu, Umar Iqbal and Jacob M. Montgomery of Washington University in St. Louis issued 55,393 trending queries across 19 topical categories over a 40 day window running from 13 March to 21 April 2026, in a paper submitted to arXiv on 13 May 2026 and revised on 1 October 2026, and report in its abstract that of the domains an AI Overview cites, nearly 30 percent do not appear in the first page of results for the same query at all. This blog read that paper when it was published and wrote up what it means for a site that ranks nowhere. Whatever the 15 October changes do to the links in an answer, the pool they are drawn from was never just the top ten, and the AI crawler reaching your server is the part of that pipeline you can still inspect.
Flow: Crawler requests the page (measurable) to Readable prose returned; Readable prose returned (measurable) to Snippet directive allows quoting; Snippet directive allows quoting (stated condition) to Indexed and snippet eligible; Indexed and snippet eligible (opaque) to Chosen as a supporting link; Chosen as a supporting link (impressions only) to Impression counted, clicks not reported.
Lantad
Published .
Every few weeks Google publishes a post saying that its AI answers will link out more, and every time the same question goes unanswered: more than what. The post of 15 October 2026 is the clearest example so far. It names five separate changes to AI Mode citations and to the links inside AI Overviews, it is specific about what each change does, and it carries no measurement of any of them. There is no before figure, no after figure, and no percentage. A site owner reading it learns that something moved and cannot learn how far.
Common questions
What changed for AI Mode citations on 15 October 2026?
Google announced five changes to how AI Mode and AI Overviews link out: more inline links surfaced higher in responses, a tested section of official links at the top of a response, news carousels in additional locations and languages, related article links made more prominent with image thumbnails, and a test that makes creator and publisher names on video results clickable to their Search profiles. Three are described as launching or rolling out and two as tests.
How much did inline links increase?
Google did not say. The post of 15 October 2026 describes substantial increases and attaches no number, no percentage and no comparison date to them. The only figure in the post is that Preferred Sources users have grown more than 7x since the start of the year, which is stated with no base count and no method, and which is about a different feature.
Is there anything a site owner has to change because of this?
No markup, file, header or setting is named in the announcement, and Google's AI features documentation, which carried a last updated date of 2025-12-10 UTC when read on 17 October 2026, names no control specific to these surfaces beyond robots.txt, the snippet directives, noindex and Google-Extended. The stated precondition for being shown as a supporting link is unchanged: the page must be indexed and eligible to appear in Search with a snippet.
Can the effect of this change be measured from outside Google?
Not directly. Search Console's generative AI performance report, read at Google's help documentation on 14 August 2026, offers impressions with no clicks, no click-through rate, no average position and no query dimension, so a change in how prominently a link is displayed cannot be separated from a change in how often it is shown. What can be measured is whether a crawler receives a readable page at all, which Lantad counts on a fixed corpus of 1,419 hostnames.
See what AI can read on your site
Run a free scan and get a graded report of exactly what AI crawlers can and cannot read, with ranked fixes.
AI visibility
AI visibility is whether AI systems can access, read, and faithfully represent your web pages, the prerequisite for being cited in an AI answer.
AI crawler
An AI crawler is an automated bot, such as GPTBot, ClaudeBot, or PerplexityBot, that fetches web pages so an AI system can read, train on, or cite what those pages contain.
Generative Engine Optimization (GEO)
Generative Engine Optimization (GEO) is the practice of structuring and publishing content so that generative AI engines can fetch it, read it in full, and cite it when they answer a question.
Google AI Mode: 132 of 1,083 home pages sent the directive that governs it, and 3 limited anything
Google's robots meta tag documentation states that nosnippet prevents content being used as a direct input for AI Overviews and AI Mode, and that max-snippet limits how much may be used. Lantad read the home page of all 1,419 hostnames in this repository's two committed corpus seed files on 7 October 2026 with no JavaScript executed. 1,083 answered HTTP 200 with HTML. 132 sent one of those two directives. 129 of them set max-snippet to -1, which Google defines as no limit at all.
Google Search profile badge: 2 of 1,093 home pages referenced it, both as data Google never asked for
Google added a guide on 16 September 2026 describing how to put a Search profile badge on a website, and it asks for a downloaded image and a link rather than any markup. Lantad captured the home pages of all 1,419 hostnames in this repository's committed corpus on 3 October 2026 and read the 1,093 that returned HTML. Two referenced profile.google.com, both inside a sameAs array on an organisation node, both using a URL form the guide does not document, and neither carried the text link the guide suggests.
Search Console's generative AI report counts impressions and nothing else
Google's generative AI performance report, read at its Search Console help documentation on 14 August 2026, offers one metric and four dimensions. There is no clicks column, no click-through rate, no average position and no query dimension, and the Search and Discover versions count an impression by opposite rules.
