Google plans to make its Discover feed adjustable with Artificial Intelligence (AI): you will tell a chatbot which content you want to see more or less often. The change affects not only readers, but also publishers and small businesses whose content is selected by feeds, search engines, and AI answers. Being discoverable increasingly means more than appearing in a results list; a system must also consider you a suitable source.
What is changing at Google
Google Discover recommends articles based on your activity across Google Search and other Google apps. According to the announcement of the new Discover feature, you will be able to adjust the feed in your own words through its three-dot menu. A chatbot-style interface will confirm your choices, describe the types of content it intends to prioritize, and let you add details.
You can then refresh the feed to apply the changes. Google also says the system will remember your preferences for later visits. The report said the feature would roll out to the Google app in the coming days, but it did not provide a country-by-country or language-by-language schedule.
Consider an everyday example: if your Discover feed contains too many stories about one topic, you can tell Google that you want less of it and more coverage of another subject. You use natural language, meaning an ordinary sentence rather than a collection of category menus and switches. The report says YouTube, Instagram, Bluesky, and X have also introduced AI-supported ways to customize feeds.
Google plans to make daily audio briefings in the Android version of Google News customizable as well. It is also expanding Preferred Sources, a feature that lets you give selected publishers more prominence in Top Stories, AI Overviews, and AI Mode. Publishers can now place an interactive Preferred Sources button directly on their websites.
More control over your feed
The customization can be useful if you already use Discover and want more focused recommendations. It does not remove the need to assess individual stories: a feed tuned to your interests does not tell you whether an article is complete, current, or reliable. Personalization initially improves relevance, not necessarily quality.
Step 1: Describe your interests precisely
- Open the three-dot menu for an item in the Discover feed inside the Google app.
- Describe the topics you would like to see more or less often.
- Review the summary in which the chatbot confirms its understanding of your preferences.
Step 2: Correct the selection
- Add more detail if Google has interpreted your request too broadly or too narrowly.
- Use the provided option to refresh the feed.
- On later visits, check whether the saved preferences actually produce more relevant recommendations.
Step 3: Choose preferred sources
- Use a Preferred Sources button on a participating website or Google’s source preference page.
- Select publishers you want to follow regularly.
- Treat the choice as extra weighting, not a guarantee that every answer will include that source.
Google said people had already selected more than 345,000 unique sources when it discussed the expansion. Earlier company studies also found that people were twice as likely to click through to a preferred source when one was available. These figures come from Google and were not independently verified in the supplied reports.
Why visibility on the AI web matters
The change affects the entire path between a question and a website. Traditional search engines present a list of results, while generative AI systems, services that compose new responses from available information, often provide a direct written answer. Source links may be included, but they do not necessarily receive the attention of a conventional search result.
This new form of discoverability has produced the term Generative Engine Optimization (GEO). It means optimizing content for generative answer systems. The idea is modeled on Search Engine Optimization (SEO), which aims to improve a website’s presence in conventional search results.
A review of automatically monitored ChatGPT searches illustrates how opaque source selection can be. The Promptwatch service tracks responses to predefined prompts across ChatGPT, Claude, and Gemini. Its data suggested that the share of monitored ChatGPT background searches containing a “site:” operator, which limits a search to a specified website, rose in early August from roughly 0.3 to 0.5 percent to 16 or 17 percent.
Those figures apply only to the prompts monitored through Promptwatch’s automation and do not establish how all ChatGPT searches behave. The technical cause also remained unclear. The author believed the search tool might use separate fields for query, recency, and domains rather than inserting the operator directly. Promptwatch later reported a major decline in the use of Reddit, but the corresponding system change could not be confirmed.
For businesses, the practical lesson is simpler than the technology: an AI service can alter how it chooses sources without publishing a ranking or a precise explanation. A company may be visible and then appear rarely in similar answers after a change. GEO measurements are snapshots, not deeds to a permanent piece of an answer.
Pros and Cons of personalized AI selection
Pros:
- More direct control – You can express your interests in ordinary sentences instead of merely hiding individual posts.
- Greater weight for trusted sources – Selecting preferred publishers can highlight their work across several Google products.
- More relevant professional information – A focused feed can give greater priority to subjects from your industry.
- Another route to small providers – Business directories and AI answers create an additional way to discover companies beyond search result lists.
Cons:
- Opaque selection – The precise weighting of interests, sources, and background searches remains largely invisible to you.
- Narrower perspective – A feed strongly tailored to existing preferences may show other relevant viewpoints less often.
- Platform dependence – Publishers and businesses depend on changes that Google, OpenAI, or other providers can make quickly.
- Limited measurement – Observations from individual tools cover only particular prompts, systems, and periods.
The public AI Observatory addresses a related lack of transparency. According to a report on its analysis of more than 24,000 chats, the platform examines real conversations with models including Claude and Gemini. The conversations were collected with user consent from seven existing datasets.
The project is intended to provide an independent complement to usage reports from OpenAI and Anthropic. Its analysis found differences between models, changes over time, and more sensitive uses than the reports from major AI providers indicated. It does not directly measure the visibility of individual websites, but it highlights a broader problem: statements from platform operators do not necessarily show the full picture of how their products are used or make selections.
What small businesses can do
Switzerland offers a concrete example through Cohaga, a company based in St. Gallen. Its ranQ directory designed for AI searches is intended to show whether and how companies appear in answers from ChatGPT, Claude, and Gemini. According to the company website, it covers more than 600,000 Swiss businesses and draws on sources including daily synchronized Swiss commercial register data and publicly available company information.
Companies can edit their profiles after verification. For a beginner, this is a sensible first step: check whether the public information about your company is correct and update the verified profile when the service allows it. That does not guarantee a mention by a chatbot, but it provides a more controllable data foundation within the directory.
Advanced users can monitor recurring, realistic customer questions across several AI services and document the differences. Tools such as Promptwatch and ranQ promise this type of visibility measurement, although their findings reflect only the prompts and systems they test. A single ranking matters less than whether company information appears accurately and consistently across several relevant questions.
A workplace example shows why this can have commercial consequences. According to the 2026 Carpathia business-to-business monitor, about 27 percent of respondents used AI for product and supplier searches. Among companies with more than 50 employees, the figure was 54 percent. The online survey ran from April to July 2026 and included 146 people, so its results describe that sample rather than every Swiss company.
AI visibility does not determine a purchase on its own. Real-time availability in an online store was important or fairly important to 96 percent of participants, while 92 percent said the same about having a personal contact. A supplier must therefore do more than appear in an AI answer; it still needs to have products available and people who can respond. The chatbot may point to the door, but the offer still has to walk through it.
What remains unresolved
For users in Switzerland, it remains unclear when the new Discover controls will become available, in which languages, and with what exact scope. The reports mention the Google app and, for customizable audio briefings, Android, but provide no separate Swiss availability details. The supplied sources also offer no detailed privacy information about how saved feed preferences will be handled.
It is equally unclear how long a preferred source or a well-maintained business profile will affect selection. Google points to higher click-through rates for Preferred Sources, while the observations involving ChatGPT show how quickly search patterns can shift. The claims do not directly conflict, but they emphasize different sides of the issue: Google highlights additional user control, while external observers expose hard-to-see changes in the background.
Customizable feeds give you more influence over what you see, but they also place selection power in systems whose rules remain only partly visible. For small businesses, AI visibility is becoming a supplement to SEO, accurate company data, and reliable service rather than a replacement for them. The unresolved risk is that platform changes can alter reach and source selection without giving affected readers or businesses a clear account of cause and effect.
Sources
- Google Discover is getting an AI chatbot-tuned feed – The Verge, 2026-08-20
- ChatGPT search now uses the site:operator at scale – Simon Willison’s Weblog, 2026-08-20
- Google gives publishers a new way to fight AI-driven traffic losses – TechCrunch, 2026-08-20
- Cohaga lanciert KI-Firmenverzeichnis – Netzwoche, 2026-08-21
- Analyse von mehr als 24.000 Chats: Diese Plattform will zeigen, was KI-Firmen nicht verraten – t3n, 2026-08-21
- Grössere Unternehmen setzen beim Einkauf stärker auf KI – Netzwoche, 2026-08-21
