AI Search Privacy: What Changes When You Ask Instead of Search

Additional resource for the lesson “Searching with AI: What Changes for Your Data” — Online Privacy course

AI search privacy starts with a small change in habit. Instead of three keywords, you ask a full question and a follow-up, and the answer comes back as one written reply rather than a list of links. This resource looks at what that means for your data: what your questions say about you, where they are kept, what your account adds to them, and how to check the answer before you rely on it.

A. Why this matters

Searching used to mean fragments: “train times Turin”. With AI answers in search engines, and chat tools that search the web for you, the natural way to ask is a sentence with context: what you are planning, for whom, and why. More context gives a better answer. That is the trade.

For your privacy, this means a question carries far more about you than a few keywords, is usually kept as history linked to your account, and may start a search before you have noticed one is happening.

None of this is a reason to stop using AI search. AI search privacy, in practice, means knowing where your questions go, deciding how long they stay, choosing what your account adds to them, and keeping your own judgement about the answers.

What the sources say

The providers’ help pages describe how AI search handles a question. Google explains that AI Mode “divides your question into subtopics” and searches for each one at the same time, a technique it calls query fan-out, and that past AI Mode searches are saved as history in your account. OpenAI explains that “ChatGPT may search the web automatically when your question would benefit from current information”, and that it “typically rewrites your query into one or more targeted queries” for its search partners, without sending them your IP address or account information.

Both companies are frank about accuracy. Google says AI Overviews “can and will make mistakes” and that AI Mode “may misinterpret web content or miss context”. Its advice is simple: “Always check important info in more than one place.”

The US Federal Trade Commission (FTC), in a note to AI companies, observed that customers “may reveal sensitive or confidential information when using a company’s models”, and that companies must keep the privacy commitments they make.

Human review, training and temporary chats have their own resources in this course, on where chatbot conversations end up and on opting out of AI training.

B. Key concepts

Five ideas about what changes when you search by asking.

AI answers in search

A summary written by AI, shown above the usual results or in a separate conversational mode: Google calls these AI Overviews and AI Mode. Chat tools such as ChatGPT and Gemini can also search the web and answer with links.

Google says AI Overviews cannot be turned off, but after a search you can select the Web filter, which “displays only text-based links without features like AI Overviews”.

Why it matters to you: An answer that looks finished invites you to stop there. The pages behind it show who is actually saying what.

Longer, more personal questions

Questions written as full sentences, with follow-ups that build on what you said before. Google notes that AI Mode handles “questions that might have previously taken multiple searches”. Behind the screen, one question can become several searches.

Why it matters to you: Context improves the answer and adds information about you. “Deposit return rules” says little. “How do I tell my landlord I am moving out in June without losing my deposit?” says what you are planning and what worries you. A chain of follow-ups says more again, because each one adds a detail to the last.

Search history

The record of your questions and answers, saved so that you can pick up where you left off.

Google is moving search history to a new setting, Search Services History, gradually; until it reaches you, Web & App Activity controls it. History is kept for a period you choose, then deleted automatically, unless you keep it indefinitely. AI Mode history lets you delete one search or all of them, though Google notes that deleted searches “may still appear in My Activity for some time”.

Why it matters to you: History is useful, and it is also a record of what you were wondering about. Deciding how long it lasts is one of the simplest privacy choices there is.

Linked to your account

When you are signed in, your questions join your account, and the account can shape the answers.

Google says its Personalized Recommendations are “based on info in your Google Account, like your profile info”, your Search Services History and other saved activity, including your general location; you can turn them off under Data & privacy › Personalization settings. In AI Mode you can also connect Gmail, Google Calendar and Google Photos; Google says “all Search services will be permitted to access Workspace and Google Photos” once you consent; you can disconnect them at any time. OpenAI says that if memory is enabled, ChatGPT “may use relevant saved memories when rewriting a search query”, and that “Device location sharing is optional and off by default” (Settings › Data controls › Location).

Why it matters to you: Separate questions, joined to one account over months, become a picture of you, and each connection lets the search see more of your life. That can be worth it, if you choose it. It is also why the account deserves protection (section H).

Checking sources

The links or citations that come with an AI answer. OpenAI’s help page says: “Select a citation to open its source”. It also warns: “Search results and citations can be incomplete, outdated, or incorrect.”

If you report a wrong answer, Google notes: “The feedback you submit includes your most recent search query and its search results.”

Why it matters to you: Opening one source, checking that it says what the answer claims and looking at its date keeps the final judgement with you.

C. A practical example: Elena’s launch week

Elena is 35 and works in marketing, where she handles client data. In the week before a client’s product launch, she used AI search on her phone between meetings, signed in to her personal account. Her questions included:

• A venue for 80 people in Turin, with the client’s name and the product’s name, not yet announced.
• A follow-up: how to tell a client that a launch date may have to move.
• At home: how to ask for a pay rise after five years at the same agency.
• In the evening: pottery classes near her flat.

On Friday she opened her AI search history to find the venue answer, and scrolled. The list read like her week: a client’s unannounced product, a work worry, her career plans, her evening routine and roughly where she lives. She was not alarmed; she had never seen it all in one place.

She also noticed that the venue answer gave a capacity of 120 people. On the cited page, that figure belonged to a different hall in the same building.

What changed when Elena looked

• She checked her agency’s policy on AI tools: client names and unannounced products now go only into the tool the agency approves.
• She deleted the client conversation and set her history to delete itself after a shorter period.
• She kept the pay-rise question: the answer helped, and she was comfortable with the trade.
• She kept personalisation on, now knowing what it drew on, and left her email unconnected.
• She opens one cited source before a fact goes into a presentation.

She still uses AI search daily; she now decides what stays and what it can see.

D. Try it yourself: read your AI search trail (15 minutes)

Four short steps, with the AI search or chat tool you use most. Nothing you find means you have done anything wrong.

Step 1 — Find where your questions are kept (4 minutes)

• If you are signed in, your questions are probably saved to your account.
• In Google, look for AI Mode history, or open My Google Activity and find Search Services History (or Web & App Activity, if the new setting has not reached you yet).
• In Gemini, look for Gemini Apps activity. In ChatGPT, searches are part of your chats, so they stay in your chat history until you delete them.
• While you are there, note whether personalisation is on, and whether any of your own apps, such as email or calendar, are connected to your search.

Step 2 — Read your last ten questions as a stranger would (4 minutes)

• Read your ten most recent questions as if they belonged to someone else, follow-ups included.
• Ask: what could a reader work out about this person’s work, home or plans? Which details did the answer actually need?
• Mark each one: fine, phrase differently next time, or delete. It is your own picture, not a score.

Step 3 — Choose how long your history lasts (4 minutes)

Pick one option; each is a legitimate choice:

• Keep it: the history is useful to you, and now you know what it holds.
• Shorten it: set automatic deletion to a period that suits you, or delete the few questions you marked.
• Pause it: turn history off for new searches, knowing that what is already saved stays until you delete it.
• Use a temporary chat for one-off questions in a chat tool, so that they stay out of your history.

Step 4 — Check one answer against its sources (3 minutes)

• Open one citation or link from a recent answer you relied on and check: does the page say what the answer says, when was it updated, and who wrote it?
• Ask the same question in a different way, or use the Web filter, and see whether the picture changes.
• No source to open? Treat the answer as a starting point, not a conclusion.

E. Videos, articles and further resources

Institutional sources first, then the help pages of the services discussed here.

Institutional sources

FTC (US) — AI Companies: Uphold Your Privacy and Confidentiality Commitments
A note to AI companies: users may share sensitive information, and privacy promises must be kept.
https://www.ftc.gov/policy/advocacy-research/tech-at-ftc/2024/01/ai-companies-uphold-your-privacy-confidentiality-commitments

AI answers in a search engine

Google Search Help — Find information in faster & easier ways with AI Overviews in Google Search
How AI Overviews work, their mistakes, the Web filter and what your feedback includes.
https://support.google.com/websearch/answer/14901683?hl=en-GB

Google Search Help — Get AI-powered responses with AI Mode in Google Search
Query fan-out, follow-up questions, AI Mode history, personalisation and connected apps.
https://support.google.com/websearch/answer/16011537?hl=en-GB

Google Search Help — Find & manage Search Services History
What the new history setting saves and how to turn it off.
https://support.google.com/websearch/answer/17024959?hl=en-GB

Google Search Help — Get started with Search Services History & Personalized Recommendations
What personalised recommendations draw on, and where to turn them off.
https://support.google.com/websearch/answer/17025248?hl=en-GB

Chat tools that search the web

OpenAI Help Center — Searching the web with ChatGPT
When ChatGPT searches, what reaches search partners, how location and memory are used, and how to open citations.
https://help.openai.com/en/articles/9237897-searching-the-web-with-chatgpt

Gemini Apps Help — Gemini Apps Privacy Hub
Where Gemini keeps your activity and how to manage it.
https://support.google.com/gemini/answer/13594961?hl=en-GB

Links checked in October 2026. If an address changes, searching for the title on the organisation’s site usually finds it.

F. The Cyber Welfare Framework: Skills, Awareness, Secure Behaviour

This resource sits on the Awareness pillar at level FL2 – Beginner: seeing what a new way of searching does with your questions, then making one choice.

Skills

• Finding where your AI searches are saved and how long they are kept.
• Checking personalisation and connected apps in the account you search with.
• Checking an AI answer against the source it cites.

Awareness

• Understanding that AI search privacy depends on how much context you put into a question.
• Knowing that one question can become several searches, shaped by your history, memory and location.
• Recognising that a confident answer is not the same as a checked one.

Secure Behaviour

• Leaving names of clients, colleagues and other people out of questions to public tools.
• Reviewing your AI search history now and then.
• Opening at least one source before an answer goes into a decision or a document.

G. Questions to sit with

• Which recent question would you phrase differently if it were read back to you in a year?
• Is your AI search history more useful to you, or more revealing about you?
• When did you last open the page behind an AI answer before acting on it?

H. What to do now

The recommendations (R) and security measures (MS) from the Cyber Welfare database that apply most directly here.

1. Your history and your sources

Steps 3 and 4 hold the choices here: how long your history lasts, and checking an answer before relying on it.

Minimum commitment: set automatic deletion on the AI search history you use most, or decide consciously to keep it as it is.

2. The account that holds your history

• R4 — Protect your account with a second factor: your AI search history lives in your account, and anyone who got into the account could read it.

Minimum commitment: check that the account linked to your AI search has a second factor turned on.

In short

• AI search turns a few keywords into a conversation, and a conversation says more about you.
• Your questions are usually kept as history and linked to your account; you can choose how long they stay.
• Personalisation, memory and connected apps let AI search draw on more of your life: each is a setting you can check.
• AI answers can be wrong: open one source before you rely on one.

Related resources in this course

• Search Engine Privacy: What Your Search History Says About You
• Alternative Search Engines: What Changes for Your Privacy
• AI Chatbot Privacy: Where Your Conversations End Up
• What Not to Share with AI Chatbots: Keep, Generalise, Share

Discover more companion resources from the online courses of the Protect Your Digital Privacy programme.