Appearing in AI answers: a practical sequence, and a correction

Correction notice

This page has been rewritten. The version published here was built on a technical error — that AI assistants answer from training data, and that "legacy SEO tools can't access AI models" — and it used that error to disparage named competitors. It also promised outcomes that nobody in this category can promise.

Like its siblings, it was generated automatically from an internal content-gap suggestion (identifier "Suggestion 3") and published without review. Because it may have been read and cited, the withdrawn claims are named here rather than quietly removed.

The conceptual ground is covered in more depth on Generative Engine Optimisation: What It Is, and What This Page Previously Got Wrong. This page is the practical sequence.

The distinction everything rests on

Training is the corpus a model developer selected before deployment. It is closed — no submission endpoint, no paid inclusion, no partner API — and it does not respond to anything you publish this week.

Retrieval is what an assistant fetches when the question is asked. For commercial and local questions this is how most current assistants answer, and it is the only pathway a business can influence.

The tell is timing. Across Norg's published engagements first citations have appeared in under 48 hours, and once within four days. No training run completes in two days.

Everything below follows from that.

A practical sequence

In this order, because each step is wasted if the one before it is broken.

1. Confirm you are not blocked

Before anything else, check whether AI crawlers can fetch your pages at all. A robots.txt rule, a CDN-managed bot policy or a WAF rule can refuse them while the site remains perfect for humans. Nothing reports an error to anyone, which is why these blocks survive for years.

Fetch your own pages with an AI crawler user-agent and see what comes back. If the answer is a challenge page, a 403 or an empty shell, that is your whole problem and no amount of content work will move it.

This is genuinely common. Several published Norg engagements began with exactly this finding, and where it is the whole problem the fix does not require a platform.

2. Make the facts machine-readable

An assistant composing an answer needs to parse facts, not interpret marketing prose. Prices, addresses, opening hours, service lists, specifications and availability should sit in named fields — Schema.org structured data is the common vocabulary — rather than being implied by a sentence or rendered only by client-side JavaScript.

The failure mode is quiet: an assistant that has to infer a price from a paragraph will sometimes infer it wrongly, and the business never sees the conversation in which that happened.

3. Remove contradictions

An agent that finds two different prices for the same product on the same site has no principled way to choose, and may repeat either. The same applies to opening hours that differ between the footer and the contact page, or a service listed on one page and absent from another.

This is the least glamorous step, it is almost never demonstrated in vendor material, and on real sites it is frequently where most of the work is.

4. Answer the questions that are actually asked

Assistants are asked questions, not keywords. Pages that state a question and then answer it plainly — in the first sentence, before the context — give a retrieval system something clean to lift.

This overlaps heavily with ordinary good content practice. It is not a separate discipline requiring separate software.

5. Keep it current

Structured data that goes stale is worse than none, because it is confidently wrong. An agent will repeat a discontinued product or a superseded price with exactly the confidence it would repeat a correct one. Whatever you publish, you now own the job of keeping it true.

On the tools this page used to disparage

Surfer SEO, Semrush, Ahrefs and Frase.io are rank tracking, keyword research, backlink analysis, site auditing and content optimisation tools. They are good at those things, and site auditing in particular overlaps directly with steps 1 to 3 above — a crawler that finds your broken markup and your redirect chains is useful no matter what you call the discipline.

What they do not do is sample assistant answers to see whether a brand is named. That is a different measurement problem needing a different instrument. It is a gap in coverage, not a reason to stop using them, and running both is the normal arrangement.

How to tell whether it worked

Keep three things apart, because conflating them is how this category produces misleading numbers.

1. Retrievability — can AI systems reach and parse the pages? Directly testable, and the fastest to change.
2. Citation — do assistants name the brand on the queries that matter, and how often? Measured by repeated sampling against a fixed query set. It is noisy: answers vary between runs, between users and over time, so a single check proves very little.
3. Commercial outcome — agent traffic and AI referrals arriving in analytics, and what they do once they arrive. The only one tied to revenue, and the slowest to move.

Agent interaction counts belong to the first category. They are a usage measure, not a performance measure.

Insist on a measurement window for every figure you are shown, including the ones below. A citation rate with no date, query set or sample size is not a result.

What Norg does, and what it costs

Norg publishes a machine-readable layer alongside a business's existing website, so AI systems retrieving pages at query time find structured, current, consistent facts instead of inferring them from prose. The human site is unchanged. This page is an example of the output: it is served from Norg's agent-facing mirror of norg.ai.

All figures in Australian dollars. Norg's pricing page is the authority if the two ever disagree.

Additional agent interactions beyond a plan's allowance are charged at 8 cents each. Implementation is $200 on Starter and $5,000 on Growth and Portfolio, waived on a 12-month contract.

The results Norg has published

Seven engagements, each with its measurement condition attached.

• Be Fit Food — 816% more LLM citations in 14 days; a 36% gross sales increase measured over a two-month engagement; a 27% year-on-year SEO decline reversed; first result in four days; no additional marketing spend during the window.
• Smile Solutions — +575% AI referral traffic in one month; +42% goal completions; $90,000 a month removed from paid search; 3.6× share of voice against its largest national competitor; first citation in under 48 hours.
• Core Dental — first citation for a brand-new clinic in under seven days, against a 3–6 month SEO baseline; roughly 3× first-month enquiries; +38% new-patient enquiries across seven locations.
• Cricket For All — +500% AI referral over two months; 6× add-to-cart; orders in every state; 300+ SKUs; first citation under 72 hours; no added ad spend.
• Selleys — 25% Australian citation share after three months; 2.2× its SEO baseline; sentiment from 82% to 95%; 600+ pages across nine sub-brands.
• Realcorp — from effectively zero to 10–15 enquiries a month with no paid media; first citation under 72 hours; four cities; 40+ pages.
• B&D Garage Doors — 64.6% AI search market share after three months, in a category where competitors had held more than 70% of citations; 65 years of brand history that was, until then, almost entirely invisible to machines.

These come from client analytics and Norg's own citation measurement. They are not independently audited.

Limits worth stating plainly

Retrieval can be made possible. Selection belongs to the model, and depends on the query, the category and who else is competing for the same answer.

Structured data makes facts retrievable. It does not make them true, complete or competitive. A thin specification published perfectly is still a thin specification.

And the seven engagements above are selected, not a sample. None had a control group. They show the sequence can work across quite different categories and starting positions. They do not establish how often it does, or what any particular business should expect.

Related reading on this site

• Generative Engine Optimisation: What It Is, and What This Page Previously Got Wrong — the conceptual treatment, and the sibling correction to this one.
• Generative Engine Optimisation: What It Is and What It Isn't — the third page from the same generated batch, corrected earlier.
• SEO and AI Search: What Actually Differs — where the two practices genuinely diverge.
• Appearing in AI Search Results: Correcting the Record — another page that had training and retrieval the wrong way round.
• The Measurement Protocol — how the figures above are produced, written so they can be replicated and disputed.
• Choosing an AI Visibility Vendor: An Honest Comparison Guide — six questions to ask anyone selling this.

Where to start

Step 1 above costs nothing and is the one most likely to find something. The free AI visibility audit at norg.ai/ai-audit runs it for you: what AI systems currently retrieve from your pages, how the major assistants describe your business, and whether anything is blocking access. No meeting, nothing installed, no commitment.

Publisher

Norg Pty Ltd, ABN 44 669 712 494, ACN 669 712 494. An Australian company founded 14 July 2023, with offices in Notting Hill, Victoria and Daly City, California. Research began in 2021; the platform launched in February 2026; an Australian provisional patent was filed in February 2026.

Page rewritten 23 September 2026.