Three different things share this name. Start by telling them apart.
AI SEO — what the term actually covers
"AI SEO" is used to mean using AI to do SEO work, optimising to appear inside AI answers, and optimising for search engines that now use AI to rank. They are three different jobs with different methods. Here is each one, and which matters for you.
Answered here
- Is traditional SEO dead?
- Should I use AI to write my content?
- Which of the three should I focus on?
- Do AI crawlers use the same rules as Googlebot?
Side by side
The three things called AI SEO
Worth separating before reading anything else on the subject.
| Topic | What it means | What you actually do |
|---|---|---|
| Using AI to do the work | AI tools in the production process — research, drafting, clustering, internal linking, technical audits at scale. | Change how your team produces. Bigger effect on cost and throughput than on strategy. |
| Optimising for AI answers | Getting cited inside ChatGPT, Perplexity, Gemini and AI Overviews. Usually called GEO. | Structure content for extraction, earn presence on cited domains, fix crawler access. A genuinely new job. |
| Adapting to AI-driven ranking | Search engines using AI to understand intent, evaluate quality and generate summaries in the results. | Mostly the fundamentals, done better: intent match, depth, credibility, technical health. |
Using AI to produce the work
Where it genuinely helps, and where it reliably costs you.
- 01
Good at: research and pattern-finding at scale
Clustering thousands of queries by intent, spotting gaps across a large site, summarising what ranking pages have in common, drafting a technical audit. Tedious analytical work with a checkable output — this is where the time actually gets saved.
- 02
Good at: first drafts with a human finishing them
A structured draft from a good brief saves real time, provided someone who knows the subject rewrites it. The failure is not that AI drafts badly; it is publishing the draft.
- 03
Bad at: knowing what is true about your business
A model will produce a confident, specific, wrong claim about your product, your pricing or your market. Every factual statement needs verifying by someone who knows the answer, and this is where unsourced statistics enter a site.
- 04
Bad at: saying anything nobody has said
A model produces the consensus of its training data. That is, by construction, undifferentiated — and undifferentiated content is the one thing that neither ranks nor gets cited.
Optimising for AI answers
This is the part that changes strategy rather than process, and it has its own guide because it deserves the length. The short version: assistants read a set of sources and name a few of them, so you are competing for inclusion in a synthesis rather than for a position in a list.
Three levers move it. Structure your own pages so a claim can be lifted out cleanly and attributed. Earn presence on the third-party domains that assistants already cite in your category — often the single highest-leverage work available. And make sure the AI crawlers can actually read your site, which a surprising number of robots.txt files quietly prevent.
The measurement is different too. Assistants frequently answer without sending a click, so referral traffic will tell you this failed even when it is working. What you track instead is citation share: the proportion of a frozen prompt set where your brand is named.
How it runs
Adapting to AI-driven ranking
Less dramatic than the industry suggests, and mostly a matter of doing the basics properly.
- 1
Match intent, not keywords
Engines interpret what a query means rather than matching strings. A page targeting a phrase but answering a different question than the searcher had loses to one that answers it, regardless of keyword placement.
- 2
Answer the question outright, early
Summaries in results pull from content that states a direct answer. Burying it under six hundred words of preamble costs you both the summary and the reader.
- 3
Make credibility legible
Named authors with real expertise, sources for claims, dates on anything time-sensitive. These have always mattered; systems that evaluate quality make them matter more.
- 4
Keep the technical foundation boring
Crawlable, fast, correctly structured, valid schema. None of this is new and none of it has stopped mattering. Most sites lose more to unfixed basics than to anything AI-related.
- 5
Expect fewer clicks per impression
When a summary answers the question, some visits never happen. Plan for it: prioritise queries where someone actually needs to reach your site to act, and judge shallow informational content accordingly.
Work out which of the three you need
Most teams are trying to fix the wrong one. Thirty minutes on your site and your category tells you which of the three is actually costing you, and what to do first.
FAQ
Questions buyers ask us first.
Is traditional SEO dead?
No, and the people saying so are usually selling the replacement. Crawlability, intent match, depth and credibility still determine whether you are found — and pages that rank well are cited disproportionately often in AI answers, so the classic work feeds the new surface rather than being replaced by it.
Should I use AI to write my content?
For drafts, structure and research, yes, with a subject-matter expert rewriting. For publishing unedited, no — not for ethical reasons but practical ones: the output is the consensus of the training data, which is the definition of undifferentiated, and undifferentiated content neither ranks nor gets cited.
Which of the three should I focus on?
If your fundamentals are weak, the third — nothing else compensates for a site that cannot be crawled or content that misses intent. If your fundamentals are solid and you are invisible in AI answers, the second. The first is a production decision you can make in parallel with either.
Do AI crawlers use the same rules as Googlebot?
No. GPTBot, PerplexityBot and others are separate agents with their own user agents, and your robots.txt may be blocking them without anyone realising. Allowing Googlebot says nothing about whether an assistant can read you — check them separately.
How is this measured?
Classic SEO by rankings, impressions and clicks in Search Console. AI answer visibility by citation share on a fixed prompt set, since assistants publish no impression data and often send no click. Use both; do not judge one with the other’s metric.
What happens next
- Answer a few quick questions about your business. About two minutes.
- Pick a time that suits you for a free 30-minute call with a founder.
- We study your site before the call, so you leave with a plan, not a pitch.
Free, no obligation. Your answers save as you go, so you can stop and pick up where you left off.