An AI search marketing strategy, sometimes called AI search optimisation, makes a business visible and accurately described inside AI-generated answers, not only in traditional search results. Infokus Marketing builds these for Australian expertise-led firms through a structured programme called Advertise to AI.
Usually because the site is turning them away, and nobody chose to. This is where the largest share of avoidable failure sits, and it is invisible in Search Console.
An Australian law firm we work with had every part of its AI visibility programme ready to go. Before starting we read the robots.txt file, which is the instruction sheet a site gives to crawlers. It was blocking nine of them, including ClaudeBot, GPTBot and Google-Extended.
Nobody at the firm had chosen that. The rule had been injected automatically at the network layer by a security service, switched on by default, and it had been working against the site for months.
The important part is what those blocks actually do, because most advice on this subject gets it backwards. There are four kinds of crawlers, and they are not interchangeable.
Training crawlers collect content for model training corpora. GPTBot, ClaudeBot and CCBot are training crawlers. Google-Extended is not a crawler at all. It is an opt-out token that makes no requests. Blocking any of these has no effect on whether a business gets cited.
Retrieval crawlers build the index an AI engine cites from. OAI-SearchBot, Claude-SearchBot and PerplexityBot are retrieval crawlers. Blocking one removes the business from that engine’s cited answers.
Agent crawlers fetch a page live because a person asked an assistant to read it. ChatGPT-User, Claude-User and Perplexity-User are agents. Blocking one refuses to load the page for a real, high-intent visitor.
Search crawlers are Googlebot and Bingbot. Never block Googlebot, and remember Bingbot feeds Microsoft Copilot and parts of ChatGPT search.
Agent and retrieval traffic is the citation-relevant majority. In a first-party study of 3,392 AI crawler hits over fourteen days, agent fetches accounted for 44.3% and retrieval 15.1%, close to sixty per cent combined, against 29.6% for training. OpenAI’s three crawlers alone were roughly 58% of all AI bot traffic.
So the common self-inflicted wound is a single blunt rule that blocks “AI bots”, removes the business from AI answers, and saves a little training bandwidth in exchange. That is what had happened to the law firm.
Two checks are worth running on any site. Open yourdomain.com.au/robots.txt and read it, looking for Disallow rules against the retrieval and agent tokens above rather than only the training ones. Then confirm the sitemap declared at the bottom of that file actually loads.
Two further things can silently override a perfect robots.txt, and neither appears in Search Console. A CDN can block AI crawlers at the edge regardless of what the file says, and Cloudflare’s managed robots.txt will set a training policy on a business’s behalf. And in WordPress, the Settings, Reading “discourage search engines” checkbox overrides everything.
Passing these checks does not establish that a business will be cited. Failing them does explain a great deal, and it is worth understanding why a website can be invisible in AI search before spending anything on content.
New first sentence: Because a model will only recommend a business it can recognise as one real, coherent organisation.
When an AI tool decides whether to name a business, it is answering a question about confidence. Does this appear to be a real, specific, coherent organisation that does the thing being asked about?
It builds that picture by reading everything it can find and looking for agreement. The website, the Google Business Profile, the LinkedIn company page, the founder’s profile, directory listings, any press.
Where those sources agree, confidence forms. Where they disagree, it does not.
Disagreement is rarely dramatic. A business name written two ways. A service list that changed on the website but not on LinkedIn. An old page still describing an offer retired last year. Two versions of an About page both live at once. Each looks harmless alone. Together they stop a model from committing to a recommendation, which is the mechanism behind how AI platforms decide which firms to recommend.
So an AI search marketing strategy usually begins with an inventory of what a business currently says about itself in every place it appears, and a decision about the one version that is correct. Consistency means agreement on the facts. Each platform can still use language suited to its readers.
Credible outside references matter for the same reason. Current professional memberships, relevant editorial, client evidence used where permission exists. The aim is accurate corroboration rather than a collection of mentions in unrelated directories.
By answering the question in the first sentence and building the page so any section survives being read alone. Content built for traditional search assumes a click, and the page does the persuading once the reader arrives. Content built for AI search assumes no click at all. The job is to be the passage the model lifts.
That changes the structure. Answer first rather than building to it. Name the business inside the answer, not only in a byline. Write sections that make complete sense in isolation, because that is how they will be read. Use the language buyers use rather than the language the industry prefers.
There is peer-reviewed evidence for this. Research presented at KDD 2024 by Aggarwal and colleagues at Princeton and IIT Delhi tested a range of content changes against live generative engines. Citing sources, adding credible quotations and adding statistics were the top-performing methods, improving source visibility by up to 40%. Keyword stuffing performed about 10% worse than doing nothing.
It also changes what is worth writing about. The highest value pieces answer real questions nobody in the market has answered well. What something costs. How one option compares to another. How long results take. These are unglamorous, and they are the questions buyers ask AI tools most often, precisely because they are awkward to ask a salesperson.
By running a fixed set of buyer questions repeatedly and recording three different outcomes separately. AI answers are not deterministic. Ask the same question three times and you can get three answers naming different businesses, so a single check tells you almost nothing.
Record whether the business is named, whether one of its pages is linked as a source, and whether the answer actually recommends the business for the buyer’s specific need. These are not the same thing. A citation to an educational article is useful. It is not a recommendation to hire the firm.
Fix the conditions and write them down. The platform, the date, the location assumptions, the session settings. Repeat in fresh sessions and save both the answers and the source links. Three runs give an initial observation. They are too small a sample to establish a dependable visibility rate, and anyone telling you otherwise is selling certainty that does not exist yet.
Progress also arrives in a rough order. Machines acknowledge before they cite. Articles get indexed, the site starts appearing in the source lists underneath answers even when the business is not named in the answer itself, and citations follow. Referral traffic from AI platforms comes last and stays small, because most of these tools pass little or no referral data.
A business measuring only the last of those will conclude nothing is happening while a great deal is. Ask new prospects how they found you, because analytics alone will not capture an AI-assisted journey.
Four stages, and the sequence is not negotiable. Infokus structures the Advertise to AI programme around Foundation, Strategy, Build and Amplify.
Foundation fixes what is broken and settles what the business says about itself. Crawler access, sitemap, schema, and one agreed description used everywhere. This is unglamorous and it is where most of the eventual result is decided. It is also where what Advertise to AI actually means becomes concrete rather than theoretical.
Strategy works out which questions matter. Not keyword volume, but the questions a buyer asks an AI tool on the way to a decision, and which of those nobody in the market currently answers well. The gap is the opportunity.
Build answers those questions properly. A small number of substantial pieces, each owning one question, each written so a passage can be quoted without losing its meaning. An AI search marketing strategy does not need volume. It needs the right five or six pieces done well.
Amplify continues the work and reviews visibility over time, running the same buyer questions across the platforms each quarter. That is what turns the programme from a hope into something that can be managed.
Firms that try to start at Build almost always waste the effort, because they are publishing into a site the crawlers cannot read or an entity the models cannot recognise. Research and measurement inform decisions throughout rather than happening once.
The firms winning here are not the biggest or the loudest. They are the ones that are legible: readable by the crawlers, consistent everywhere they appear, and clear enough in their writing that a model can quote them without ambiguity. That is mostly a discipline problem rather than a budget one, and an AI search marketing strategy is largely a series of small corrections applied consistently rather than a large investment applied once.
It is also early. The category is open in a way traditional search has not been for fifteen yearsMost Australian expertise-led firms have not started, and some of the ones that have are blocked at the first step without knowing it. A firm that gets this right this year is competing against a very small field.
That will not stay true. It is true now.
The two checks in this article take about ten minutes and will tell you whether your site is visible to AI tools at all. If you would rather have someone run the full picture, the Infokus AI Visibility Audit tests your business with ten buyer questions across four AI platforms, three passes each, and shows you where you appear, where you do not, and who is being named instead.
Does blocking AI crawlers protect my content?
Only from model training, and at a cost most businesses do not intend. Infokus Marketing finds that blanket “block AI bots” rules usually remove a business from AI answers while saving very little, because training crawlers and the retrieval and agent crawlers that produce citations are separate things governed by separate tokens. Decide the training question deliberately, then make sure the retrieval and agent crawlers are still allowed through.
Is AI search optimisation different from SEO?
AI search optimisation builds on SEO rather than replacing it. Google has stated that existing SEO practices apply to AI Overviews and AI Mode with no special additional markup required, so useful content, accessible pages and clear internal linking still do the underlying work. An AI search marketing strategy adds a further question: whether the business is named and accurately described in the answers buyers receive.
How long does an AI search marketing strategy take to work?
Expect the first signals within a quarter and a readable picture by the second. Foundation work is quick to complete and slow to show, because crawlers have to return and reindex before anything changes. Infokus Marketing sees acknowledgement first, as pages appear in the source lists underneath answers, then citations, then any referral traffic. A business measuring only the last will conclude too early that nothing is happening.
The Referral Problem Referrals are the foundation of most professional services practices —
What “Advertise to AI” actually means — and what it doesn’t Advertise to
Most business owners have moved past the question of whether to use AI.
Client overview Galvin-Rowley Executive is a people-first executive recruitment and advisory business. Its
