Here is the uncomfortable shift nobody in your industry wants to say out loud: a growing slice of your future customers will never see your website at all. They will ask an AI assistant a question, read the three-sentence answer it gives them, and act on it. If your business is not inside that answer, you do not exist in that moment, no matter how well you rank on page one of Google.
For twenty years, the game was simple. You optimised pages, earned links, climbed the rankings, and waited for the click. The click is now optional. Google AI Overviews, ChatGPT search, Claude, Microsoft Copilot and Gemini increasingly read the web, digest it, and hand the user a synthesised answer with a couple of cited sources. The question for every Perth business owner is no longer just "where do I rank?" It is "when an AI explains my market, does it understand me well enough to recommend me?"
That is where a small, almost invisible file called llms.txt enters the picture. It will not fix a broken business. It will not paper over a thin offer. But it does something quietly powerful: it removes the guesswork an AI has to do about who you are, what you sell, and why you are the right answer. In a world where machines summarise you before a human ever meets you, that clarity is worth far more than another blog post nobody reads.
What is llms.txt?
llms.txt is a plain-text Markdown file you place at the root of your website, so it lives at yourdomain.com/llms.txt. Its job is to give large language models, the engines behind tools like ChatGPT and Gemini, a clean, curated map of the most important content on your site. Think of it as a one-page briefing written specifically for machines, in language they can read without tripping over your navigation menus, cookie banners, pop-ups and JavaScript.
The idea was proposed in September 2024 by Jeremy Howard, the Australian researcher behind fast.ai and co-founder of Answer.AI. His observation was practical. Modern AI models work within a limited context window, a fixed amount of text they can hold in mind at once. A typical web page is mostly not content. It is layout, scripts, tracking, related-post widgets and three different calls to action. When an AI tries to understand your business from that soup, it wastes most of its attention on noise and often gets the substance wrong.
llms.txt fixes that by doing the editing for the machine. You hand it a short document that says, in effect: here is who we are, here are the eight pages that actually matter, and here is one line on what each of them covers. It is the difference between dropping a stranger in the middle of Perth CBD with no map, and handing them a single index card with the five addresses they actually need.
It is worth being precise about what llms.txt is not. It is not a robots file that blocks or grants access. It is not a sitemap listing every URL. It is not a magic ranking switch. It is a curated, human-readable summary that points an AI to your best material and frames it in plain words. That simplicity is the entire point.
Why this matters for Perth businesses right now
The behaviour has already changed. Search the way you actually use these tools yourself. When you want a recommendation, you increasingly type a full question into an AI assistant rather than two keywords into Google, and you trust the answer it gives back. Your customers in Joondalup, Osborne Park and Fremantle are doing exactly the same thing, often without realising the mechanics have shifted underneath them.
This is the discipline people are starting to call Generative Engine Optimisation, or GEO, sometimes Answer Engine Optimisation. It is the natural successor to traditional SEO. The goal is no longer only to rank a blue link. It is to be the business the machine names when it composes the answer. Ranking gets you considered. Being legible to the model gets you cited.
Here is the part that should sharpen your attention. AI engines are biased towards sources they can parse cleanly and trust quickly. A Perth business with a tidy, unambiguous description of its services has a real edge over a larger competitor whose site is a tangle of clever animations and vague brand-speak. For once, the smaller, sharper operator can win on clarity rather than budget, particularly with a Perth digital marketing consultant sharpening the message. llms.txt is one of the most direct ways to claim that clarity.
Ranking gets you considered. Being legible to the machine gets you cited.
llms.txt, robots.txt and sitemap.xml: who does what
These three files get lumped together because they all live at the root of your domain and they are all plain text. They do completely different jobs, and confusing them is the fastest way to get this wrong.
The mistake almost every WA business is making
The default move right now is to do nothing and assume that strong Google rankings automatically carry over into AI answers. They do not. Rankings and AI citations are related but separate. An AI can rank you well in classic search and still misrepresent you in a generated answer, because it built that answer from whatever fragments it could scrape and understand, not from a clean brief you controlled.
The second mistake is the opposite extreme: spending heavily on a visually stunning website that is, to a machine, almost unreadable. Heavy JavaScript frameworks, content loaded only after interaction, text baked into images, endless pop-ups. A human with patience can navigate it. An AI on a tight context budget gives up and moves to the competitor whose site it can actually digest.
The third mistake is vagueness. Walk through your own homepage and count how many sentences a stranger could use to say precisely what you do, for whom, and where. On most WA business sites the answer is close to zero. It is all "elevating brands" and "tailored solutions" and no plain statement of fact. Humans tolerate that fog. Machines penalise it, because they cannot summarise what was never clearly stated.
How to build your llms.txt: the framework
You do not need a developer for the thinking, only for the final upload. The hard part is the editorial discipline of deciding what genuinely matters. Work through these steps in order.
Audit your real money pages
List the pages that actually explain and sell your business: your core service pages, your about page, your key location pages, your two or three best resources. Ignore the rest. If a page would not help an AI recommend you, it does not belong in this file.
Write a one-paragraph identity statement
In plain English, state who you are, what you do, who you serve and where. "All Marketing is a Perth digital marketing consultancy in East Perth WA, helping small and medium WA businesses with SEO, Google Ads and content." No adjectives doing the work of facts.
Structure the file in Markdown
Start with an H1 of your business name. Add a blockquote with your identity statement. Then add H2 sections such as Services, Locations and Resources, each containing bulleted Markdown links, with a short note after each link describing what that page covers.
Write the notes for a machine, not a brochure
Each link gets one concrete sentence. "SEO services for Perth businesses, including technical audits, local SEO and content." Specific, literal, scannable. This is the text the model leans on when it decides whether to cite you.
Add an optional section and a fuller version
Use an H2 called Optional for links that are useful but not essential, so a model short on context can safely skip them. For deeper engines, publish a companion llms-full.txt containing the full clean text of your key pages.
Host it at the root of your domain
Upload the file so it resolves at yourdomain.com/llms.txt. On WordPress you can do this with a plugin or by dropping the file in your site root. Confirm it loads in a browser as clean plain text, not as a styled page.
Maintain it and measure the lift
Review the file each quarter and whenever your services change. In Google Analytics 4, watch for referral traffic from AI tools such as ChatGPT, Claude and Gemini. It will start small. The trend is what matters.
That is the whole job. An afternoon of clear thinking, a text file, and a quarterly review. The reason most businesses will not do it is not difficulty. It is that writing a genuinely clear description of yourself is harder than hiding behind jargon, and far more revealing.
What a finished file actually looks like
It helps to see the shape of the thing, because the simplicity surprises people. A working llms.txt for a Perth consultancy reads almost like a contents page. The top line is a heading with the business name. Beneath it sits a single blockquoted sentence of identity: a Perth digital marketing consultancy in East Perth helping WA businesses with SEO, Google Ads and content.
Then come the sections. Under a Services heading, a short bulleted list of links: the SEO page, with the note "technical and local SEO for Perth businesses"; the Google Ads page, with "search and performance campaigns managed locally"; the content page, with "blog, web copy and GEO content". Under a Locations heading, the East Perth and wider WA service pages. Under Resources, the two or three articles that genuinely demonstrate expertise. Finally an Optional heading for the nice-to-have links a context-starved model can skip without losing the plot.
That is the entire document. No styling, no cleverness, no marketing varnish. Every line is a fact a machine can lift straight into an answer. The discipline is in the restraint: the temptation is to include everything, and the whole value of the file evaporates the moment it stops being a curated shortlist and becomes another sprawling index.
A Perth example of the framework in action
Meridian Fitouts and the invisible enquiry
To show how this plays out, picture a typical Perth business we will call Meridian Fitouts, a mid-sized commercial fit-out firm based in Osborne Park. It is a composite, built from patterns we see again and again across WA service businesses, and the numbers here mirror what genuinely shows up in the field rather than any single real client.
Meridian ranked respectably for "office fit-out Perth" and had a handsome, animation-heavy website. Yet when its director asked ChatGPT and Gemini to recommend commercial fit-out companies in Perth, Meridian was never named. Competitors with plainer sites were. The AI simply could not work out, quickly enough, what Meridian did and where.
The fix took an afternoon. A clear identity paragraph. A llms.txt file listing eight pages: three service pages, the Osborne Park location page, two project case studies, the about page and contact. One literal sentence on each. Nothing clever, just clear.
Within roughly two months, Meridian began appearing in AI-generated shortlists for Perth fit-out queries, and GA4 logged a small but steady stream of referral sessions from AI assistants, climbing from zero to a few dozen qualified visits a month. Modest numbers, but every one of them was a buyer who had asked a machine for a recommendation and been handed Meridian's name. That is the difference between being on the web and being in the answer.
The objections, handled
"The big AI engines do not officially use llms.txt yet, so it is a waste of time."
Partly true, and it still misses the point. Major search crawlers have not confirmed llms.txt as a ranking input, and adoption is uneven. But a growing list of AI-native tools and developer platforms already read it, your own AI integrations can use it today, and standards reward the early and well-prepared. The file costs an afternoon and nothing to host. Waiting for universal confirmation is how businesses always arrive late to a shift that was obvious in hindsight.
"I already have a sitemap, so the machines have what they need."
A sitemap lists every URL with zero context. It tells a model that two hundred pages exist, not which eight explain your business or why you are worth recommending. Handing an AI your sitemap is like answering "what does your company do?" by reading out the table of contents. llms.txt is the summary the sitemap never provides, and the two work together rather than competing.
"Publishing a plain summary just hands my competitors a map of my business."
Everything you would list already sits on your public website for anyone, competitor included, to read. llms.txt does not expose anything new. It simply states clearly what was always there. The businesses that win in AI search are not the ones hoarding a secret, they are the ones a machine can understand fastest. Clarity is the advantage here, not obscurity, and your competitor copying your honesty does not make you any less the cited source.
The long game for Perth businesses
None of this replaces the fundamentals. You still need a clear offer, a fast and well-built website, genuine authority and the technical SEO foundations that have always mattered. llms.txt is not a shortcut around weak marketing. It is an amplifier for businesses that already have something clear to say and simply have not made it legible to the machines now standing between them and their customers.
The deeper opportunity is the discipline itself. The act of writing a llms.txt forces a clarity most businesses avoid for years. Once you can describe yourself in one honest paragraph, that clarity flows into your content, your ads, your sales conversations and your whole online marketing. The file is the artefact. The thinking is the prize.
Find out where your business is invisible to AI
In a focused 60-minute Marketing Growth Diagnostic Session, we will pressure-test how Google and the AI engines actually see your business, and map the fastest moves to get you cited. $175 + GST, no jargon, no obligation.
Book Your Session