AI Mode
AI Mode is Google's conversational search mode. Instead of ten blue links, the user gets one continuous answer assembled from several sources and can keep asking follow-up questions. Sources are cited in the answer, but clicking through is optional and most people never do it.
In short
| What it is | Google's conversational search mode with a model-generated answer and cited sources |
| Who it matters for | Brands whose customers ask about comparisons, recommendations and how-to questions, which covers most B2B and e-commerce |
| What it replaces | Nothing. It sits next to classic search as an additional mode |
| How you win in it | By being a cited source, not by holding the top link position |
| How it is measured | Impressions and positions on long question-shaped queries, citations in assistants, referral sessions from assistant domains |
| When to ignore it | For purely navigational and transactional brand queries, where users hardly ever use the mode |
How AI Mode works
Classic search matches a query against pages and returns a ranked list. AI Mode first breaks the query into several sub-questions, retrieves sources for each and assembles one answer from them. The user then keeps asking, and the context of previous questions carries over.
Queries are longer and written as sentences
In classic search people type two or three words. In conversational mode they write full sentences, often with conditions and context. Instead of "ppc agency" they ask "which agency should I pick if I spend two million a year and I am expanding into Germany". Content written for short head terms does not answer that.
The answer is assembled from several sources at once
The model does not pick one winning page, it combines several. That means a site that would rank on page two in the classic list can still make it into the answer, provided it holds a clear, quotable answer to one of the sub-questions.
How it differs from AI Overviews
AI Overviews is a summary that appears above the classic results on an ordinary query. The user did not ask for it and can only follow up by moving into AI Mode. AI Mode is a separate mode the user enters deliberately and holds a conversation in. For content, that creates a difference in length: AI Overviews pulls short factual statements, AI Mode can take longer explanations too, because the model has room to develop an answer across several rounds of questions.
What the model takes as a source
The model does not work with your website directly. It works with what sits in its index and with what retrieval returns at the moment of the query. For Google, the source is the ordinary Googlebot index. For other assistants (ChatGPT, Claude, Perplexity) two separate layers are in play: the indexing crawler that pulls content into the knowledge base, and the crawler that fetches a page while answering a specific question. Either can be blocked on its own, often without anyone in the company knowing.
The click is optional
The source link is visible in the answer, but the user already has what they came for. That is why these queries routinely produce high visibility and zero traffic. It is not a measurement error, it is the nature of the channel.
From our own practice: what we watch on conversational queries
On conversational queries we look at different numbers than in classic SEO. The key one is the ratio between impressions and clicks: on a long question-shaped query a page routinely holds a top position and gets no click from it, while the very same page sits deep on a short head term. That is not a contradiction. The long query matches the structure of the content; on the head term the whole site is competing.
Sessions from assistants arrive with a long lag
The second thing to plan for is the lag between visibility and traffic. Search impressions move first, sessions from assistant domains arrive in single units and months later. By default analytics counts them as ordinary referrals, so they need a channel of their own; without it they disappear in the report entirely.
The practical consequence: a company expecting traffic from this channel will read the result as a failure even when everything is working. The value of the channel is not the session count, it is the brand being mentioned at the moment someone is deciding.
The crawler you do not know you are blocking
Technical availability is therefore verified directly: a request to the server carrying the crawler's user agent, and a check of the response code. A block rarely comes from a decision in robots.txt. It comes from the hosting firewall or the CDN, a layer above the CMS that nobody normally looks at. A split between two bots of the same provider is common: the crawler answering a live question passes, the indexing crawler gets HTTP 403. The site then stays usable inside an individual answer but never enters the model's knowledge base.
The practical conclusion: on these queries, optimizing for clicks is the wrong goal, and technical availability has to be verified directly rather than assumed from the configuration.
Common mistakes
- Judging AI Mode by clicks. Zero clicks at position one looks like failure in a report, while it is success in a different metric.
- Writing only for short keywords. A conversational query carries conditions and context that keyword-shaped copy does not address.
- Burying the answer at the end of the article. The model lifts a clear statement, ideally right under a heading. What sits in the tenth paragraph rarely makes it in.
- Blocking language model crawlers. A site a crawler cannot reach cannot be cited. It usually happens by accident at firewall level, not in robots.txt.
- Leaving no single machine-readable source of truth about your brand. When consistent structured data is missing, the model fills the gap from elsewhere.
- Expecting a fast effect. Weeks to months pass between a content change and a citation in an answer, because the model has to take the content into its index first.
How to measure visibility in AI Mode
Google gives no dedicated report yet, so it is measured indirectly. The process we use:
- In Search Console, filter queries longer than six words or containing a question mark. That is the closest proxy for conversational queries.
- Track position and impressions on them, not clicks. The goal is impression growth while holding a position inside the top five.
- Add citation counts from Bing Webmaster Tools, which reports AI Performance.
- In Google Analytics, watch the ai-assistant channel, or build your own segment on chatgpt.com, gemini.google.com, perplexity.ai, claude.ai and copilot.microsoft.com. The numbers will be small, the point is the trend, not the volume.
- Once a month, verify crawler availability directly by requesting the site with their user agents. Checking robots.txt is not enough.
- Manually check a sample of queries inside AI Mode and record whether the brand is mentioned and in what wording.
It helps to keep a simple monthly sheet with four numbers: how many conversational queries produced impressions, the average position on them, the citation count from Bing Webmaster Tools and the number of sessions from the ai-assistant channel. One row a month is enough, and after six months it shows a trend that an ordinary SEO report would lose.
What to do about it in practice
The order of steps that makes sense for most sites:
- Verify availability. Go through the language model crawlers and confirm none of them receives a 403 or a 429. This is the one step without which the rest is pointless.
- Put the answer up front. The first two or three paragraphs under a heading have to carry the full answer, one that still makes sense taken out of context.
- Write headings as questions. A structure that mirrors how people actually ask reads far better to a model than topical headings.
- Add your own numbers. Original data from accounts, measurements or internal tools is the strongest reason for a model to cite you, and the one thing competitors cannot copy.
- Deploy structured data. FAQPage, Organization and, on term pages, DefinedTerm. It will not do the content's job, but it makes comprehension easier.
- Align the facts about your brand. The same numbers on the site, in profiles and in directories. A contradiction gets resolved by the model on its own terms, usually not in your favor.
Related terms
See also AI Overviews, zero-click search, AEO, GEO, llms.txt and citation share.
Frequently asked questions
Will AI Mode replace classic search?
Not so far. It sits alongside it and people use it mainly for more complex queries. Navigational and purchase queries still return the classic list of results.
Why do we hold a high position and get no clicks?
Because the user received the answer inside the interface. On conversational queries this is normal and does not indicate a problem with the page title or description.
How do you get into the answer?
With a clear answer up front, a question-led heading structure, machine-readable data and your own numbers that nobody else has. Original data is the strongest citation factor.
Does schema help?
It helps comprehension. Structured data tells the model what is on the page, but on its own it guarantees nothing. Without a good answer in the text, schema is useless.
How long until changes show up?
Weeks, similar to organic search. Bing tends to be faster, because other assistants draw from it and indexing can be accelerated with the IndexNow protocol.
Can visibility in AI Mode be bought?
Ad formats in conversational search are only now being introduced and have nothing to do with source citations. Citations cannot be paid for, the model selects on content and availability.
How much traffic does it actually bring?
A handful of sessions per month at the start, even with good visibility. Months of work on content and availability usually pass before the first assistant session arrives. Anyone expecting volume comparable to search will be disappointed. The value lies in the brand being mentioned at the moment someone is deciding.
How we can help
Visibility in AI search engines is one of our services, from technical availability for crawlers through structured data to content that can actually be cited. Details are on the AI visibility agency page.