Surface marks show where New Reward checks visibility. They are not partner badges or outcome guarantees.

Back to Field guide
Field guide
Published |8 min read

Google is building the tool inside AI Overviews. Formula pages are next.

Google confirmed generative UI is rolling into AI Overviews. On September 7, Search Engine Roundtable documented an interactive character-counter widget on "character counter" queries. Google already showed mortgage calculators and study tools. Here is the operator split: public-math tools get replaced; first-party proof does not.

James Brady portrait

By James Brady

Chief AI Officer, Product & AI Operations

AI OverviewsGenerative UIGEO

Google can now generate the interface, not just the summary.

That is not a blogger take. It is Google's own product language. On August 19, 2026, Google's back-to-school Search post said generative UI "have launched globally in English in AI Mode and have started rolling out as well in AI Overviews." On September 7, Search Engine Roundtable documented the next obvious step: an interactive Character Counter widget inside AI Overviews for character-counter queries — the kind of page that used to earn the click by being the tool.

Sources: 5 new ways to level up your learning with Search, Google, August 19, 2026; Google AI Overviews Get A Character Counter Tool, Search Engine Roundtable, September 7, 2026; Google Expands Generative UI Beyond AI Mode Into AI Overviews, Search Engine Journal; Google's Generated Interfaces Could Compete With Tool Pages, Search Engine Journal; Generative UI: A rich, custom, visual interactive user experience for any prompt, Google Research, November 18, 2025; Generative UI: LLMs are Effective UI Generators, arXiv:2604.09577; AI features and your website, Google Search Central.

The short version

  • Google's Aug 19 post: generative UI is live globally in English in AI Mode, and "have started rolling out as well in AI Overviews."
  • Same post's example is an interactive visual in an AI Overview for a query like "ph scale," then a follow-up into AI Mode.
  • Sep 7, Search Engine Roundtable: AI Overviews now loads an interactive Character Counter on character-counter queries. Observation from Damir B via Barry Schwartz. Not a Google blog post.
  • Google Research already framed the destination: an "infinite catalog" of ephemeral interfaces generated on the spot, instead of picking from a finite library of apps and pages.
  • In Google's own preference study (arXiv HTML opened this fire), generative UI beat the top Search result in 90% of LMArena-prompt pairwise comparisons, and in 73.5% on an information-seeking set. Lab study. Generation speed ignored. Google graded its own feature.
  • Eligibility for supporting links did not invent a new markup requirement: indexed, snippet-eligible. No special AI file.
  • The operator split is simple. Tools that are public math (counters, converters, generic calculators) compete with Google's own generated UI. Tools and pages that need your private data, your live inventory, your signed-in workflow, or your dated proof still have a job.
  • Monitoring will not separate "summary AIO" from "tool AIO." Search Console's generative AI report does not label interface vs paragraph. Monitoring ≠ execution.

What Google actually said

From the August 19 back-to-school post, after the pH-scale example:

Generative UI capabilities have launched globally in English in AI Mode and have started rolling out as well in AI Overviews.

That is the line. Not "coming someday." Started rolling into the default results-page surface.

The same post also puts practice quizzes in both AI Overviews and AI Mode, free, global English. Study-aid pages are already in the blast radius. Google framed the launch around school. The capability is not limited to school.

Google Research, November 18, 2025, when generative UI landed in AI Mode:

Generative UI is a powerful capability in which an AI model generates not only content but an entire user experience.

And the paper's closing frame, quoted on the research blog and in the arXiv HTML:

…an infinite catalog, where the right ephemeral interface is generated on the spot tailored for their need.

Ephemeral is the word that matters for publishers. The tool is not a competitor site you can outrank next month. It is generated for that query, then gone.

At I/O 2026, Google's Search post said generative UI would be "available for everyone in Search this summer, free of charge," and described assembling "interactive visuals, tables, graphs or simulations" in real time. The August 19 post is the AIO-side receipt.

What operators saw this morning

Search Engine Roundtable, September 7:

Google will now give you an interactive widget within AI Overviews for when you are looking for a simple character counter website. Websites that offer character counting service features are in trouble.

Barry Schwartz's framing is blunt on purpose. Weather, dictionary, movies, mortgage calculators, emojis — Google has eaten utility SERPs before. The new part is that the widget can be generated, not only hard-coded as a OneBox.

This is observation. One query class. One reporter. No Google product blog confirming the character-counter widget by name. Treat it as a field note that the Aug 19 rollout is landing on non-study utility queries, not as a universal rule for every calculator on the web.

Search Engine Journal's Matt Southern had already drawn the publisher line when generative UI entered AI Overviews: summaries compete with pages people read; generated interfaces compete with pages people use.

What Google's own study actually measured

We opened the paper HTML (arXiv:2604.09577). Do not launder this into "90% of searchers prefer Google's tools in the wild."

What the study did:

  • Pairwise human preference between formats, including generative UI vs the top Google Search result for the prompt.
  • Raters saw pre-cached results. Generation time was not a factor.
  • Main prompt set: 100 prompts sampled from LMArena (chatbot-style). Info-seeking set separately.

What Table 2 style pairwise wins show in the HTML we opened:

  • On the LMArena set, generative UI vs Website (top result): 90% preference for generative UI.
  • On the information-seeking set: 73.5% for generative UI vs top Search result (19% the other way in the matrix SEJ and the HTML both surface).
  • Human-expert custom sites still led overall. Generative UI was "at least comparable" to expert sites in 50% of cases (abstract).

Honesty on the number: Google built the feature, ran the study, and published the preference rates. Lab, not live Search Console. Chatbot-heavy prompts inflate the gap. Speed was excluded, and the research blog itself says generation can take a minute or more. Use the figures as "Google believes generated interfaces beat the top blue link in controlled preference tests," not as your traffic forecast.

What did not change

From Google's AI features guide:

To be eligible to be shown as a supporting link in AI Overviews or AI Mode, a page must be indexed and eligible to be shown in Google Search with a snippet, fulfilling the Search technical requirements. There are no additional technical requirements.

And:

You don't need to create new machine readable files, AI text files, or markup to appear in these features.

So: no llms.txt theater, no special schema that buys you a generated-widget exemption. If your page is a public formula dressed as a lead magnet, generative UI is a product competitor sitting above the results. If your page is the only place a model can verify your price, your hours, your named clinicians, your inventory, or your signed-in action, you still own something Google's ephemeral UI cannot invent from public math.

The operator split (do this week)

1. Inventory formula tools vs proprietary tools. List every page that earns traffic because it is a tool. Mark each one: public math (unit converter, character counter, generic amortization) vs private data (your rates, your schedule, your SKU availability, your patient/portal workflow). The first column is exposed. The second column is the job.

2. Stop pitching "build a free calculator for links" as strategy. A calculator designed only to attract links now competes with an interface Google can spin for the query. If you still build a tool, build it on data only you have, and put the answers in crawlable text — not only inside the widget.

3. Keep first-party answer pages boring and specific. Prices in text. Service areas in text. Named people. Dated proof. Hours. Booking path. That is what survives both quieter citations and generated widgets.

4. Baseline the queries, not the vibes. For every tool URL in column one, note the head queries and whether an interactive widget now appears in AI Overviews. Date the check. Do it again in two weeks. You want a log, not a screenshot argument.

5. Do not expect Search Console to name the weapon. Google's generative AI performance reporting (separate Field Guide when that pack goes live) tracks impressions in AI features. It does not, as of pages opened this fire, tell you "this impression was a generated calculator." A click drop on a converter page is a clue, not a labeled cause.

Honesty block

  • Google did not publish a Sep 7 post titled "character counter in AI Overviews." That observation is Search Engine Roundtable.
  • Google did not publish a query-type list for when generative UI triggers in AI Overviews. Full AIO availability timing was "coming weeks" in the Aug 19 / SEJ coverage — we have not seen a Google "fully done" notice this fire.
  • Google did not say formula tool sites will be demoted in classic results because of generative UI.
  • The 90% / 73.5% figures are lab preference wins from Google's paper, not live CTR.
  • Pew's 8% vs 15% click figures appear in SEJ's tool-pages piece; we did not open the Pew study this fire — do not repeat those percentages in social cuts.
  • Apple clarified over the weekend that site rules for Applebot-Extended are not considered in ranking for Apple Search (About Applebot, published date on the page: September 4, 2026). Training opt-out ≠ ranking. Separate surface. Mentioned here so nobody merges it into this Google story.
  • We did not open X. We did not re-auth X. No pay-per-use.

Why this is the New Reward argument

Cloud agents — and Google's own Search UI — cite and now build against who and what they can see.

A citation was never an asset you owned. A free converter was never a durable moat either. What you can own is the set of conditions underneath both: indexed pages, snippet-eligible, first-party facts a generated widget cannot reconstruct from public math, and a proof loop that ships fixes instead of dashboard screenshots.

Monitoring tells you the SERP changed. It does not inventory your formula tools, rewrite the proprietary ones, or put prices in text. Monitoring is not execution.

Ready to turn this into a next step?

Book the appointment if you want the live walkthrough. Run the score if you want the fast read first.

Book the AI-visibility demo Get your AI-visibility score

Where this fits

This is part of how New Reward improves search and AI visibility. See the SEO, AEO, and GEO offering or read how search and AI visibility work together. Related Field Guides on the live site: A Google spam demotion doesn't stop at rankings, Why good brands disappear from AI search, GEO measurement is noisy, so build a proof loop, ChatGPT site tools / WebMCP.

Cloud agents cite who they can see. Google can now build the tool without leaving Search. What you control is still the first-party page underneath.

James Brady portrait

James Brady

Chief AI Officer, Product & AI Operations

James Brady writes from the operating-system side of New Reward: product direction, AI workflows, verification, and the proof boundaries that keep visibility work honest.

Start here

Ready to turn this into a next step?

Book the appointment if you want the live walkthrough. Run the score if you want the fast read first.

Where this fits

This is part of how New Reward improves search and AI visibility. See the SEO, AEO, and GEO offering or read how search and AI visibility work together.

Prefer New Reward in Google

Keep reading

More from the Field Guide

Field guide

7 min read

Google just called manual fact-check of AI content “critical.” Titles and schema count too.

On October 1, 2026, Google Search Central updated its guidance on using generative AI content on your website. The docs now say generative models predict a likely sequence of words from training data, so outputs may contain inaccuracies (hallucinations), and that it is critical to manually fact-check and review all AI-generated content for accuracy and trustworthiness before publishing. That review also applies to metadata that can appear in Search results — title elements, meta descriptions, structured data, and image alt text. Google did not publish a new ranking formula, a citation rate for AI Overviews or AI Mode, or an end date for the September 2026 spam update. Conditions and proof still beat another dashboard seat.

GoogleGEO

Read the guide
Field guide

7 min read

OpenAI just shipped always-on Dots. Proactive research is read-only — your pages still have to be fetchable.

On September 29, 2026, OpenAI launched Dots — always-on agents in ChatGPT. Help Center pages opened this fire say a dot can look for useful updates before you ask (“proactive research”) by reading permitted connected sources and saving private notes. Those research tools cannot send messages, change content through plugins, or control a browser or computer. Cloud browser use is a separate capability for assigned work. OpenAI did not publish ranking formulas, citation rates, or a promise that background research will open your first-party URL. Conditions and proof still beat another dashboard seat.

ChatGPTGEO

Read the guide
Field guide

7 min read

Google just made AI Mode monitoring free for everyone. Your pages still have to be the source.

On September 28, 2026, Google Search VP Robby Stein said AI Mode’s info monitoring — previously for Ultra and Pro subscribers — is rolling out to everyone globally. People can ask Search to watch topics across sites, forums, social posts, real-time data, and the Shopping Graph. Google did not say how sources are chosen or whether every update still carries web links. First-party pages that stay fresh are the condition. A dashboard watching your brand is not the same job as being the page a buyer’s monitor can ground.

AI ModeGEO

Read the guide