The agency versus in-house question has a different answer for AI search than it had for SEO, and most of the advice floating around has not caught up. It is still arguing about whether you can afford a full time marketer, when the real question is which specific tasks belong where. Because here is the uncomfortable truth from inside the industry: some AEO work is so close to your business that an agency will always do it worse than you, and some of it is so infrastructural that building it internally is a waste of money.
This is the task by task split. We are an agency, so you know which way our bread is buttered, but the honest version of this answer wins us better clients than the salesy version would, and several things on this list we will actively tell you not to pay us for.
What the work actually consists of
Answer engine optimisation breaks into five kinds of work, and they have very different skill profiles. There is access work: making sure AI crawlers can reach your site, which is firewall and CDN configuration. There is facts work: publishing the verifiable claims assistants quote, from licensing to pricing to response times. There is content work: pages that answer real buyer questions well enough to be cited. There is citation work: earning presence on the directories, review platforms and listicles that assistants lean on when building shortlists. And there is measurement: running a fixed panel of buyer prompts on a schedule and tracking whether you get named more often than baseline. If any of that vocabulary is new, our AEO explainer covers the foundations.
What belongs in-house, permanently
The facts. Nobody outside your business knows your year founded, your licence numbers, your actual response times, your real pricing and your service area boundaries the way you do. When we audited 500 Australian homepages, the average business published fewer than two of the six facts assistants quote. Fixing that requires no skill an owner or office manager lacks, just a decision to stop being vague. An agency can tell you which facts are missing. Only you can decide to publish your prices.
Reviews. The review ask happens at the moment of customer happiness, and only your team is in the room for it. Agencies can build the routine, the link, the timing and the template, but the habit lives with the people doing the work. Every business that has tried to outsource the actual asking has ended up with either nothing or something worse than nothing.
Subject matter expertise. The experiences, project stories, real numbers and edge cases that make content citable come from inside. The businesses that get quoted by assistants are quotable because they said something specific. An external writer can shape it, but cannot invent it, and the AI written sludge currently flooding every industry is what invention without knowledge looks like.
What is rarely worth building in-house
Measurement infrastructure. Proper AI visibility tracking means running a fixed prompt panel across ChatGPT, Perplexity and Google’s AI surfaces on a schedule, recording citation rates, and holding the yardstick still for months. The reason this needs infrastructure rather than an intern with a spreadsheet is volatility: when we ran identical queries through Google AI Mode seconds apart, the citations overlapped only 24 per cent of the time. Single checks are noise. Rate measurement across repeated runs is signal, and building that pipeline internally for one business is buying a factory to make one chair. The volatility data is in our answer volatility study.
Cross market pattern knowledge. An agency running prompt panels across dozens of clients sees which changes moved citation rates and which did nothing, across categories, months before any public playbook says so. We know ChatGPT checks accreditation registers by name before recommending trades because we watched it do so across client categories. A single business cannot see those patterns from inside its own account.
The technical layer, usually. Schema architecture, crawler access diagnosis, llms.txt and site structure are skills your web developer partially has, and the gaps are cheaper to rent than to grow. The 13 per cent of Australian sites silently blocking AI crawlers mostly belong to businesses whose IT provider never knew there was something to check.
The real cost comparison, with Australian numbers
The comparison most articles run, salary versus retainer, is the wrong one, because a full time AI SEO hire is not the realistic in-house option for most Australian small businesses. The realistic option is existing staff time plus tools plus learning curve.
The agency side is knowable. When we surveyed 42 Australian AI SEO agencies in August 2026, only 19 per cent published pricing, but among those that did, the range ran from about $750 to $6,800 a month, with most engagements landing between $1,000 and $3,000. Our own published plans run $1,800 to $4,500 plus GST. The full market breakdown is in what AI SEO services cost in Australia.
The in-house side for a small business is roughly: a capable marketing generalist salary in Australia runs $75,000 to $110,000 plus super if you hire, or five to ten hours a week of an existing person if you assign. Tools for tracking and research add $100 to $500 a month. The hidden line is the learning curve, because this field changes monthly and the person learns on your time.
The honest arithmetic for most businesses under about $5 million revenue: a hybrid costs less than either pure option and outperforms both. In-house owns facts, reviews and expertise. An agency owns measurement, technical and citation work. Above that size, or where marketing is genuinely core to the business model, building internal capability with an agency in an advisory lane starts to win.
Two configurations from the real world
Abstract splits become clearer with examples, so here are two configurations we see working, drawn from the businesses we deal with.
The first is a regional trade business, seven staff, one owner who does the quoting. Their working split: the office manager owns the facts and the review routine, fifteen minutes a day, because she is already the person customers deal with at the happy moment. The owner records a voice memo after interesting jobs, which becomes the raw material for case study content. An agency, in this case us, owns the technical layer, the prompt panel, and the citation work, at the lower end of the retainer range. The measurable outcome across two quarters was citation rate on their panel moving from under 5 per cent to the high teens, with the first AI referred enquiry, a customer who said ChatGPT suggested them, arriving in month four. The configuration works because each side is doing the part it is structurally best placed to do, and neither is pretending to be the other.
The second is a professional services firm, about thirty staff, with a marketing coordinator already on payroll. Their split runs the other way: the coordinator runs the prompt panel monthly, owns all content production with partner input, and manages the profiles. External help was scoped down to a one off technical project, the audit, schema, crawler and register alignment work, plus a quarterly advisory call. Total external spend was a fraction of a retainer, and appropriate, because the expensive recurring parts were already covered by a salary the firm was paying anyway. The mistake this firm almost made, and the reason the example earns its place here, is that they were initially quoted a full retainer that duplicated their coordinator’s job. The agency that won the work was the one that scoped itself smaller.
The pattern across both: the split follows existing capability, not company size. Ask what your business already pays for that touches this work, and buy only the gaps.
The decision in four questions
Do you have anyone with five spare hours a week? If genuinely no, the in-house half of the hybrid fails regardless of intent, and you need an agency doing more of the lifting plus a hard look at why nobody in the business can spare five hours for how customers find you.
Is your category contested in AI answers? Run five buyer prompts through ChatGPT and Perplexity. If competitors are being named and you are not, the citation and measurement work has urgency that favours renting speed. If nobody in your category shows up yet, the field is open, the DIY foundation moves you fast, and you can defer the retainer. Our DIY audit establishes this in two hours.
Is there a public register in your industry? Licensed trades, health, legal and financial categories get verified against registers before assistants recommend them. The alignment work between your register entries, your site and your profiles is detail heavy and mostly one off, which suits a short agency project rather than either a retainer or internal trial and error.
Can you hold a yardstick still for six months? The businesses that see results measure the same panel monthly and change one layer at a time. If your internal culture changes direction quarterly, an agency’s process discipline is half of what you are paying for, whether anyone says so out loud.
The hybrid, mapped out
Month one: agency runs the audit, fixes crawler access and schema, sets the prompt panel baseline. In-house publishes the six facts and starts the review routine. Months two to four: agency builds the citation layer, directories, the sources assistants actually quote in your category, and iterates content structure. In-house feeds real project stories, numbers and answers into that content. Ongoing: agency measures monthly and reports citation rates against baseline; in-house keeps facts current and reviews flowing. At ninety days, the rate either moved or the plan changes. That gate belongs in writing before anything starts, and any agency that resists it is telling you something.
Frequently asked questions
What does an AI SEO agency cost in Australia compared to hiring?
Most agency engagements run $1,000 to $3,000 a month. A marketing hire capable of owning this runs $75,000 to $110,000 plus super, plus tools and a live learning curve. For businesses under roughly $5 million revenue, a hybrid of agency infrastructure plus internal facts and expertise beats both pure options on cost and outcome.
Can my existing SEO person just learn AEO?
The foundations transfer, and a good SEO will pick up the facts and schema layers quickly. The parts that transfer least are measurement design, because AI answer volatility breaks the habits rank tracking builds, and cross market pattern knowledge, which only comes from seeing many accounts. Read GEO vs AEO vs SEO together and you will both know where the seams are.
What should I never outsource?
Your facts, your review asks and your expertise. Nobody external can decide to publish your pricing, be present at the moment a happy customer would leave a review, or supply the specific project details that make content worth citing. Agencies that offer to fully replace those things end up publishing generic content under your name, which is the opposite of what gets quoted.
How do I know if an agency is actually doing the work?
Monthly citation rates against a fixed baseline panel, a work log of what shipped, and the ninety day review gate agreed in writing. If reporting is screenshots of single answers, that is anecdote, not measurement, because identical queries overlap only about a quarter of the time in what they cite.
Is now too early to invest in this at all?
The foundation layer is free to cheap and also improves conventional search, so there is no scenario where publishing verifiable facts and fixing crawler access is premature. The retainer question depends on whether your category is contested in AI answers yet, which five prompts will tell you in ten minutes.
What if I start in-house and switch to an agency later?
That order works well, with one warning: keep your prompt panel and baseline from day one, even as a simple spreadsheet of dated runs. A business that arrives with six months of measurement gets a better plan from any competent agency than one arriving with a feeling that things have not improved.
