AI SEO: what changes when the answer comes before the link.

AI SEO describes two different jobs. One uses AI to produce SEO work faster. The other adapts content for search results and assistants that answer a question before showing links.

Both are real, and they matter differently. The first changes how much work you get through in a week; the second changes whether that work still produces enquiries you can count.

This guide focuses on the second because it changes discovery and reporting. It also explains where AI can reduce production time without removing research, verification or human approval.

The owner decision: does this affect the pipeline?

Check the share of qualified enquiries that begins with non-branded search. If most work comes from referrals, outbound sales or people searching the business by name, AI search is not the first priority. If buyers discover the business by asking general questions, it deserves a measured response.

Do not start with an AI SEO tool. Start with Search Console, a fixed list of buyer questions and the CRM source data. The owner needs to know whether non-branded discovery is declining, shifting or still producing qualified enquiries.

The four terms and what they mean

People also search for what is AI SEO called now, which tells you how unsettled this is. Four terms, heavy overlap:

AEO and GEO describe nearly identical practice. Choose one term internally and keep the reporting consistent. The mechanics are covered in What is answer engine optimisation?.

What AI search does to your numbers

Pew Research Center found that 18% of Google searches produced an AI-generated summary in its March 2025 measurement. In the same research, published July 2025, users clicked a traditional search result on 8% of visits when a summary appeared, against 15% when one did not. Roughly half the onward clicks, on the same query.

That is population-scale and it is over a year old, and it says nothing about how exposed your own account is. Google has since launched dedicated Search Generative AI performance reports in Search Console. They provide a dedicated view of generative AI visibility alongside the overall web report, but currently show impressions only. You can see whether your pages appeared in AI Overviews, AI Mode and generative AI features in Discover, but not the clicks, enquiries or revenue those appearances produced.

That makes exposure measurable, but it does not settle the commercial question. A rise in impressions with flat clicks can still reflect a broader keyword footprint as well as interception, and the new report cannot separate those explanations.

We tried to settle it properly on our own client accounts and could not, for a reason worth knowing: on one Australian training provider, 221 of the 243 queries that held steady across a twelve-month gap currently carry an AI Overview. That is 91%, which leaves 22 queries to compare against and no usable control group. The full working is in What is answer engine optimisation?

For your own account, use the new report as the visibility baseline, then compare the same period against non-branded clicks and qualified enquiries. Impressions tell you that a page appeared; they do not show whether that appearance moved the pipeline.

How we use AI in production

The other half of AI SEO, and where the practical decisions are. We run this daily inside our AI search visibility work, so this section is description rather than prediction.

Drafting from live query data, not from a topic list. Every piece starts from what people are already typing, pulled from Search Console and keyword APIs, rather than from a content calendar someone brainstormed in a workshop. AI turns that query set into a structured draft quickly. The query data carries the value; the drafting speed is a convenience.

Daily conversion reads on ad accounts. Google and Meta accounts are structured by course or by role and tracked through to the CRM. AI reads the conversion data every day and flags where money is going to things that do not convert. Budget moves weekly, and a person makes the call. That is our Google Ads and Meta ads management.

Funnel-leak flagging. AI watches where applicants drop out of a form or a page sequence and surfaces the step. We change one thing at a time and count the lift, which is the whole of conversion rate optimisation. One education client’s course-page conversion moved 2.45× after a single on-page change, from 2.27% to 5.57%.

Answer-shaped rewriting at scale. Putting a direct answer in the first 60 words of 200 pages is exactly the kind of work AI does well and humans do resentfully.

Faster follow-up. Custom automations acknowledge the enquiry, provide the agreed next step and keep the lead warm until a staff member can connect personally. The aim is to reduce avoidable delay without pretending automation has replaced the conversation.

Common to all five: AI does the production and pattern-spotting, and a senior person decides. On regulated accounts, that division of labour is the only one we have found that holds up.

What this does not include

Three things people assume belong on that list.

Publishing unreviewed drafts. Every page is read and signed by a senior editor before it ships. Drafts arrive faster than they used to; the approval bar stays where it was, and the approval step is where the quality comes from.

AI-generated volume as a strategy. Publishing 200 thin pages because it now costs almost nothing is the most common failure of the last two years. Answer engines and conventional rankings both reward depth on a subject, so 12 pages that fully answer a question outperform 200 that half-answer it. Spend the savings on deeper research rather than a bigger publishing schedule.

Automated keyword selection. It causes the most expensive failures we see, so it gets its own section below.

What we tell clients to expect from it

AI production compresses the drafting stage of a content programme by something like 60–70% of the hours. Research, verification, editing, and the compliance review take as long as they always did, and on regulated accounts those stages are most of the work.

So the realistic gain is more output per month at the same quality bar, or the same output with far more research behind it. We generally choose the second, which is why this article carries first-party Search Console figures and a keyword study rather than another explainer.

Are AI SEO tools worth buying?

People searching this term also search AI SEO tools, best AI SEO tools and AI SEO free, so it is worth answering directly.

Most products currently marketed as AI SEO tools fall into three groups.

Rank trackers with a new label. They report the same positions they always did, with an “AI visibility” tab bolted on. Some now check whether an AI Overview appears for a query, which is useful. Many do not, and simply rename existing metrics.

Content generators. They draft at volume. So does a general-purpose assistant, usually better, and without a subscription tied to a template library.

Citation monitors. They query assistants on a schedule and record whether your brand is named. This is the only new category here, and it is the one worth evaluating, because the measurement problem is real and manual checking does not scale past a few dozen questions.

Before buying anything, do the free version. Google Search Console covers the divergence analysis at no cost. A spreadsheet of 10 questions your buyers ask, run manually across ChatGPT, Perplexity, and Google’s AI mode once a quarter, gives you a citation baseline in about 20 minutes. Do that twice, and you will know exactly what you would be paying a tool to tell you, and whether it is worth it.

Buy the tool when the manual version has proved the measurement matters to you and the volume has outgrown a spreadsheet. Doing it in that order is cheaper than the reverse, and the volume case arrives less often than vendors suggest.

Where AI gets it wrong

Three failures we hit often enough to plan around.

It invents the specifics answer engines reward

Answer engines reward concrete claims: fees, durations, codes, dates, eligibility. AI drafting produces those claims fluently and sometimes wrongly.

For Australian training providers this is a compliance problem. ASQA’s marketing and advertising obligations, which now sit under the information and transparency requirements of the Standards for RTOs 2025, apply to every claim about course outcomes, fees and eligibility, in every ad and on every page. A confidently invented subsidy eligibility line is a regulatory finding, not a typo.

Every claim of that type on a client page gets checked against the source document by a person before it ships, with no automated shortcut.

The tools drift semantically, and confidently

A live example from our own research, on 11 August 2026. We ran a keyword-ideas query seeded with “answer engine optimisation” through a major SEO data provider. It returned “decomposer examples”, “logline examples” and “xml examples”. Seeded with “recruitment agency marketing”, it returned casting agencies and HubSpot partner listings.

The tool returned a clean, well-formatted, useless list, and a team working quickly would have built a content plan on it. We discarded the whole pull and sourced topics from People Also Ask and related-search data instead.

The lesson generalises past that one tool: these systems fail quietly, by returning plausible output. Every automated list gets a human read before it becomes a decision.

It cannot tell you what a term is contaminated by

A volume number counts how many people type a phrase each month. Working out who those people are takes the SERP.

rto marketing looks like a natural target for an agency serving Australian training providers. Check the live SERP and the People Also Ask box asks “What does RTO stand for in Australia?” and “How to become an RTO”. Related searches are rto marketing course, pdf, certification. Those are people wanting to become an RTO or to study marketing, not RTO owners buying it.

recruitment agency marketing carries 320 searches a month in Australia. Its related searches are Marketing recruitment agencies Sydney, Melbourne, Perth. That SERP serves people hiring marketers.

Both terms would look like wins in a keyword tool and produce traffic that never converts. Reading the SERP takes two minutes and no AI currently does it for you reliably.

Does Google penalise AI-written content?

AI SEO Humanizer is a live related search on this term, which means a lot of people are worried about detection. Most of that worry is aimed at the wrong thing.

Google’s published position is that it rewards helpful content regardless of how it was produced, and penalises content produced primarily to manipulate rankings. The test is purpose and quality rather than production method.

In practice, pages get hurt for being thin, duplicative, unsourced, or published at a volume no one could have verified. AI makes all four of those cheap, which is why the correlation exists and why it gets misread as a penalty on the tool.

So “humanising” a draft to defeat a detector solves the wrong problem. A detector-proof page that says nothing still says nothing. The work that helps is the work that would have helped anyway: a specific claim with a source behind it, a real example, a number from your own account, an opinion someone could disagree with.

The concern is legitimate in one place. A page that carries a byline promises the reader a person stands behind it, and that promise is editorial rather than algorithmic. It is why every page we publish is read and signed by a named editor.

What to do this quarter

Five steps, ordered, each doable without an agency.

1. Establish whether you are exposed. Split branded from non-branded queries in Search Console. If most of your enquiries trace to people who already knew your name, AI search is a 2027 problem for you and you can stop here.
2. Run the honest version of the divergence check. Take queries you ranked for 12 months ago and still rank for. Compare CTR then and now, within position bands, on that fixed set only. Anything measured across your whole account confuses growth with loss.
3. Put a direct answer in the first 60 words of your 10 most important pages. Highest-return hour available in this discipline.
4. Use AI for production, never for verification. Draft with it, then check every fee, date, code, and eligibility claim against the source. In regulated categories, have the person who owns the compliance risk sign it.
5. Read the SERP before you commit to a keyword. Two minutes per term. It is the cheapest error prevention in marketing.

Notice that none of the five steps requires AI. The cost of producing something plausible has fallen close to zero, which leaves verification, judgment, and a point of view worth citing as the scarce inputs. Buy those first, whatever you spend on tools.

Frequently asked questions

What is SEO for AI?

SEO for AI, usually called answer engine optimisation or AEO, means structuring content so AI systems can extract it, trust it and cite it when composing a direct answer. It works alongside conventional SEO rather than replacing it, because AI answers are assembled largely from pages that already rank.

Can ChatGPT do SEO?

It can do parts of the production work well: drafting, restructuring pages around a question, generating metadata, summarising query data. It cannot verify a factual claim, read a search results page for buyer intent, or carry the compliance risk on a regulated claim. Treat it as fast production with mandatory human verification.

What is AI SEO called now?

The terminology has not settled. AEO and GEO both describe optimising for AI-composed answers and mean nearly the same thing. LLM SEO is a narrower version aimed at assistants specifically. AI SEO remains the umbrella term and stays ambiguous between “using AI for SEO” and “optimising for AI search”.

Will SEO be replaced by AI?

No, though the reporting will change before the discipline does. AI answers still need sources, and those sources are drawn largely from pages that rank conventionally. What changes is that a portion of demand gets satisfied without a click, so click-based measurement understates the value of the work.

What should I look for in an AI SEO agency in Australia?

Ask three questions. What first-party data do you have on how AI search has affected your other clients, and what did it show? Which parts of your process are AI-produced and which are verified by a person? And how do you measure success when the answer appears without a click? An agency that cannot answer the third question is selling conventional SEO with a new label. Our answer: a quarterly citation baseline on a fixed question set, the divergence check on queries you already rank for, and self-reported attribution on the enquiry form, because the click disappears but the enquiry does not.