The first time I used AI Mode properly, I asked it a question I already knew the answer to.
Not to test it. To see who it would credit.
It gave me a clean, four-paragraph answer, cited three sources, and not one of them was the page that ranks first for that query. The page that ranks first has held position one for two years. It has the backlinks, the domain age, the works.
AI Mode simply did not need it.
That is the whole story in one screenshot, and it is why I have stopped treating AI Mode as a feature bolted onto search and started treating it as a second surface with its own rules.
Here is what is actually different, and what I have changed in my own work because of it.
Quick Info
The Short Version
Traditional search answers a query by ranking documents. AI Mode answers a question by assembling passages.
That single distinction produces almost every practical difference between the two.
Ranking documents rewards the page that best covers a topic overall. Assembling passages rewards the paragraph that most cleanly answers one specific part of a question, wherever that paragraph happens to live.
In traditional search you compete page against page. In AI Mode you compete paragraph against paragraph.
Which means a modest site with six genuinely excellent passages can appear in answers that its domain would never win on the blue links.
Difference One: One Query Becomes Many
Traditional search takes your query, more or less as typed, and matches it against an index.
AI Mode breaks the question apart first. Ask it to compare two email platforms for a small business on a tight budget and it will quietly run several searches at once: pricing for each, deliverability, list limits, migration effort, reviews from small teams.
Google calls this query fan-out. The practical consequence is that you are no longer optimising for the question the user typed. You are optimising for the sub-questions the model decided to ask on their behalf.
This is why I now plan content around question clusters rather than head keywords. If someone in my niche asks a big question, I want to own four of the eight small ones underneath it.
The keyword tools will not show you those sub-questions, because nobody types them. Ask AI Mode the big question and read what it chose to cover. That list is your brief.
Difference Two: Position Stops Being the Prize
On the blue links, position determines almost everything about your click-through rate. It is a steep, well-documented curve.
In AI Mode there is no position. There is a citation or there is nothing, and the citations do not correlate with organic rank nearly as strongly as any of us assumed they would.
I have client pages sitting at position seven or eight that get cited regularly, and pages at position two that never appear. The difference is always structural. The cited pages have headings that read like questions and answers that stand up on their own.
So I have changed what I report. Rank still goes in the deck, because clients care about it and it still drives real traffic. But next to it now sits a much cruder metric: how often does this page turn up as a source when I ask the questions it should be answering?
It is manual and unglamorous. It is also the closest thing to ground truth I have found.
Difference Three: The Conversation Continues
Traditional search treats every query as a fresh start. AI Mode remembers the last three things you asked.
That gives you a genuine opportunity, and it is one almost nobody is planning for.
When a user asks a follow-up, the model reaches for sources that can support the next step, not just the first one. A page that answers the question and then answers the obvious next question is far more useful to it than a page that stops at the headline.
In practice: stop ending articles at the answer. End them at the decision. What does the reader do with this, what goes wrong, what does it cost, what is the alternative.
I add a short section near the end of most posts now that I think of as the follow-up block. It exists purely to answer the thing a curious reader would ask next.
Difference Four: What Traditional Search Still Owns
I want to be careful here, because there is a lot of noise suggesting classic SEO is finished. It is not, and acting as though it is will cost you money.
Traditional search still owns navigational and transactional intent almost entirely. People searching for your brand, for a specific product, for a login page, or for a local service to hire today are still clicking links.
It also still owns the crawl. Nothing gets cited in an AI answer that could not be found, rendered and understood by the same infrastructure that has always powered search. Broken canonicals, slow servers and blocked resources hurt you on both surfaces.
The correct posture is layered, not split. One site, one technical foundation, content structured so it works as a ranked document and as a set of quotable passages.
What I Actually Changed
Five things, in the order I did them.
That last one takes thirty minutes and has changed my content plan more than any keyword tool I pay for.
The Budget Question
People ask me how to split spend between the two. My answer is that the question is slightly wrong.
There is no separate AI Mode budget, because there is nothing to buy. You cannot bid on a citation. What you can do is spend the same content budget differently: fewer, longer, better-structured pages instead of more thin ones.
If you want a number, the shift I have made is roughly this. Where I used to publish four new posts a month, I now publish two and rebuild two existing ones. Total output unchanged, total spend unchanged, and the rebuilt pages have been the ones that started appearing in AI answers.
Refreshing is cheaper than writing and it compounds on a page that already has some authority. It is the most underrated line item in SEO and it always has been.
Frequently Asked Questions
Not replacing it, sitting alongside it. Both are live, and most users move between them depending on the question. Navigational and transactional queries still go through classic search overwhelmingly.
No. You need better-structured content. Answer-first headings, self-contained passages, named sources and specific numbers help you on both surfaces.
Partly. Search Console's generative AI reports show impressions from AI surfaces by page, but not queries, clicks or which passage was quoted. Supplement it by manually asking the questions you want to own.
Yes, both for the traffic it still delivers and because being findable and trusted feeds the same pipeline AI surfaces draw from. It just no longer guarantees a citation.
Still useful for discovery, crawl priority and trust. Their correlation with getting cited is weaker than their correlation with ranking, so treat them as necessary rather than sufficient.
With structure, not volume. Take your ten best pages, make every heading a question, make every answer stand alone, and add one real statistic per section.
Before You Go
If you take one thing from this, take the paragraph test. Pick your most important page, delete everything except one heading and the paragraph under it, and ask whether it still answers something useful.
If it does not, that is your next hour of work, and it is worth more than a month of link building.
I keep the templates and the question-logging sheet I use for this in my resources section, along with my SEO and AI search course and the tools I actually pay for.
And if you would rather hand the whole thing over, you know where to find me.
Two surfaces, one site, Tariq
I'm a marketing consultant, entrepreneur and content creator. I help businesses grow through practical marketing, websites, SEO, content and AI.
More About Tariq →