How AI Overviews choose which Melbourne businesses to show
Google Business Profile vs AEO: which moves the needle first for local businesses
A client sent me a quote last month: $2,400 to “implement llms.txt for AI search readiness”. The file they were quoting for is a plain text list of links. It takes twenty minutes to write if you know your own site, and about an hour if you don’t. Here is what it is, what it actually does, and how to write yours.
What llms.txt is
It is a markdown file you put at yoursite.com.au/llms.txt. Inside it you write one line about what your business does, then a list of links to the pages an AI assistant would need to answer questions about you — services, pricing, contact, the two or three articles that actually explain your work.
Jeremy Howard of Answer.AI proposed the format in September 2024. The reasoning is dull and correct: a normal HTML page buries its information under navigation, cookie banners, scripts and a footer with forty links. A model reading that page spends most of its budget on furniture. A markdown index hands it the furniture-free version.
The v2 revision of the proposal, published after two years of real use, added a second half that matters more than the index file itself: serve a clean markdown copy of each important page at the same URL with .md on the end, so /pricing/ also answers at /pricing.md. Then point at it from the HTML with a rel="alternate" link tag.
What it does — and what it doesn’t
Let’s kill the sales pitch first. No search engine has said llms.txt influences rankings. Google has not committed to reading it. Publishing one will not put you in AI Overviews, and anybody who tells you it will is selling something.
What it does have is real adoption on the machine side. OpenAI, Anthropic and Google all publish llms.txt files for their own developer documentation. Documentation platforms like Mintlify generate one automatically. Chrome’s Lighthouse now audits for it under its agentic browsing checks — which means it has crossed from proposal into something a browser tool expects to find.
So the honest position: it is cheap insurance with a plausible upside, not a ranking lever. Twenty minutes of writing against the chance that an agent fetching your site next year gets a clean answer instead of a scraped mess. If your website budget is tight, answer engine optimisation work on your actual page content will move more than this file will.
There is a second reason to bother, and it has nothing to do with search. Writing the file forces you to name the ten pages on your site that answer a real question. Most business sites cannot produce that list without an argument. Two of the last five we did this for ended with the owner deleting pages rather than linking them, which was the correct outcome.
Writing one in twenty minutes
Open a text editor. The format is four things in order: an H1 with your name, a blockquote summary, optional plain paragraphs, then H2 sections containing bullet lists of links in the shape - [Title](URL): one-line note.
Here is a working example for a fictional Richmond physio clinic:
# Bridge Road Physiotherapy > Physiotherapy clinic in Richmond, Victoria. Sports injuries, post-surgical rehab and clinical Pilates. Open Monday to Saturday, no GP referral needed for private patients. ## Services - [Sports injury physio](https://example.com.au/sports-injury/): assessment, treatment and return-to-sport programs - [Post-surgical rehab](https://example.com.au/rehab/): knee and shoulder reconstruction, hip replacement - [Clinical Pilates](https://example.com.au/pilates/): small group and one-on-one, reformer available ## Practical - [Fees and health funds](https://example.com.au/fees/): $95 initial, $80 follow-up, HICAPS on site - [Book online](https://example.com.au/book/): live availability - [Contact and parking](https://example.com.au/contact/): 2 Bridge Rd Richmond, street parking after 6pm ## Optional - [Blog archive](https://example.com.au/blog/)
Three rules from writing these for real sites. Keep it under about fifty links — the point is a shortlist, not a sitemap. Put the answer in the note, not just a label, because $95 initial, HICAPS on site is useful and learn about our fees is not. And anything under an ## Optional heading is explicitly droppable, so park the archive material there.
Upload it to your web root. WordPress will not serve a file you drop in the media library at the right path, so put it in the site root directory by FTP or your host’s file manager, or use a plugin that registers the route. Then load yoursite.com.au/llms.txt in a browser and check it renders as plain text rather than downloading.
The part most people skip
The markdown copies. An index of links is only half the proposal — the other half is that each linked page should also be available as clean markdown. That is a bigger job, and on a WordPress site it usually means a plugin or a small function that renders post content through a markdown converter at a .md route.
Whether it is worth doing depends on how much your business lives or dies on documentation. A software company with a developer API: yes, do it, your users’ coding agents are reading those pages daily. A dental practice in Camberwell: skip it. The index file is enough, and your effort is better spent making the human pages answer questions directly, which is the same work that gets you quoted by ChatGPT optimisation anyway.
What to do instead, if you only have one afternoon
I get asked to add llms.txt to sites that have no H1, no schema markup and a services page written in 2019. That is the wrong order. The file points at pages; if the pages are thin, you have built a very tidy index of nothing.
Rough priority for a Melbourne small business site, most valuable first: make each service page answer its question in the first fifty words; add LocalBusiness and FAQ schema; fix the pages that give a different answer to Google than they give a human; then write llms.txt. That last step takes twenty minutes and the first three take a fortnight, which tells you where the value sits.
One more thing worth checking while you are in there. Open your site in a browser with JavaScript disabled and see how much text survives. If your service descriptions are rendered by a page builder script, a plain fetch gets an empty shell — and that is what a lot of AI crawlers get, because most of them do not run JavaScript. No index file fixes that. It is the single most common reason a Melbourne business with a good-looking site gets ignored by AI answers, and we see it about once a fortnight in AEO agency in Melbourne audits.
If you want a second opinion on which of those your site actually needs, our AI SEO in Melbourne audit covers all four and costs less than the quote at the top of this article.
Frequently asked questions
Does llms.txt help my Google rankings?
No. Google has not said it reads the file, and no search engine treats it as a ranking signal. It is a machine-readability convenience for AI agents fetching your site, not an SEO tactic.
Where exactly does the file go?
At the root of your domain, so it loads at yoursite.com.au/llms.txt. Subdirectory versions are allowed in the v2 proposal — /docs/llms.txt covers everything under /docs/ — but most businesses only need the root one.
How is it different from robots.txt?
robots.txt tells crawlers which URLs they may not fetch. llms.txt tells an AI reader which URLs are worth fetching, and gives it a short description of each. One restricts, one recommends. They do not replace each other.
Do I need one if my site is small?
A five-page site gains almost nothing, because a model can read all five pages cheaply anyway. It becomes useful past roughly thirty pages, or when your important information is scattered across service pages, a blog and a FAQ.
Should I pay someone to write it?
Only as part of a larger job. On its own it is a twenty-minute task you can do yourself with the example in this article. If an agency quotes four figures for llms.txt alone, ask what else is in the scope.
Can I write it in Chinese as well as English?
Yes — the format has no language rules, and bilingual sites usually list both language versions of each key page with notes in the matching language. That is standard on the sites we build through our bilingual web design work.
Melmel — a Melbourne AI studio
We build and optimise websites so that Google, ChatGPT, Perplexity and Google AI Overviews can read them and quote them. Melbourne web design and development, AI SEO and AEO from $1,000 a month, and a $199 whole-site AI audit credited in full against your first month if you go ahead. English and Chinese.
AEO agency in Melbourne · ChatGPT optimisation · AI SEO in Melbourne · answer engine optimisation · bilingual web design
Call 0473 034 777 · email kan@melmel.com.au · our Melbourne services · Suite 704, 566 St Kilda Rd, Melbourne VIC 3004.
