7 Things That Make AI Understand Your Website
Our previous article, What Is GEO?, explained why Google has you and ChatGPT does not. This is the hands-on version (llms.txt, schema markup, FAQ): the moment an AI crawler lands on your site, what it actually sees, what it is looking for, and why so many Hong Kong SME websites get one glance before it leaves. 7 things, each explained in terms of what it is, why AI needs it, how much you can do yourself, and how WKH's Get You Found handles it. Treat it as 1 checklist and tick the boxes one by one.
The 7 things that make a website readable to AI: 1. Don't block AI crawlers in robots.txt; 2. Add an llms.txt that explains what you are in plain language; 3. Add Organization, FAQPage and Article schema, 3 types of structured data; 4. Give every page a direct answer of 60 to 80 words; 5. State the author and date on every piece of content; 6. Keep company name, address and phone consistent across the web and set sameAs; 7. Ask AI once a month in a customer's voice and record the result. All free, most doable within 1 day; WKH's Get You Found does the lot at launch.
What Does AI Actually See When It Crawls Your Site?
What an AI crawler sees is not your lovely design; it is a pile of HTML tags, text and a few specific files, and it uses these to decide what you are and whether you can be trusted.
People read a website top to bottom, looking at images and words. An AI crawler goes to 3 places first: robots.txt in the root directory (am I allowed in?), sitemap.xml (which pages exist?), and, if present, llms.txt (what are you?). Only then does it enter each page and read the heading hierarchy (H1, H2), the body text, and the structured data inside <head>.
The problem with many Hong Kong SME websites is that the entire homepage is one long image plus a few adjectives: no H2s, no FAQ, and the only company information is a WhatsApp button. To a person it may look rather nice; to AI it is close to 1 blank sheet. It cannot read "what you do, who you help, what it costs", so naturally it does not cite you.
The 7 things below turn that blank sheet into 1 page whose key points AI can read.
Thing 1: Don't Block AI Bots in robots.txt
robots.txt is the first door an AI crawler meets at your site; if it is shut, nothing else matters.
What it is: 1 plain text file in the root directory that tells crawlers which areas they may and may not enter.
Why AI needs it: OpenAI's GPTBot, Anthropic's ClaudeBot, Perplexity's PerplexityBot and Google's Google-Extended all check this file first. If you Disallow them, they will not read any of your content.
Common trap: Older "anti-scraping" templates, or certain security plugins, block every non-Google bot by default. Many business owners have no idea their site has shut AI out.
How much you can do yourself: Open yourdomain.com/robots.txt and look for any of the names above followed by Disallow: /. If you find one, delete it. 5 minutes.
How Get You Found handles it: At launch, we write a robots.txt that explicitly Allows the mainstream AI bots and links it to the sitemap.
Thing 2: llms.txt — Explain What You Are in Plain Language
llms.txt is a company profile written for AI, in plain text and plain language, placed in the root directory.
What it is: A simple standard proposed by llmstxt.org: 1 markdown-format text file whose first paragraph says what you are, followed by a list of important pages with a one-line summary each.
Why AI needs it: A large language model has limited context when reading 1 website; a well-organised summary lets it grab the key points quickly instead of guessing page by page.
In all honesty: Google's official AI optimisation guide states plainly that llms.txt and special schema are not required for Google's AI features, and Google's John Mueller has said publicly more than once that Google does not use it. But crawlers on the ChatGPT and Perplexity side have been recorded reading it. In short: no use for Google, possibly useful for other AI, and the cost is close to zero.
How much you can do yourself: Write 20 lines in Notepad and upload it to the root directory. 1 hour. You can use WKH's own /llms.txt as a format reference.
How Get You Found handles it: Generated automatically from your site content and updated every time you add 1 article, so you never have to remember.
Thing 3: 3 Types of Schema Markup (Structured Data)
Schema structured data is a machine-readable set of tags that tells AI directly: "this section is company information", "this section is a Q&A", "this section is an article".
What it is: A vocabulary defined by schema.org, placed in JSON-LD format inside each page's <head>. People cannot see it; only crawlers can.
The 3 types SMEs need most:
| Schema | Where it goes | What it tells AI |
|---|---|---|
| Organization | Site-wide (usually the homepage) | Company name, logo, address, phone, social accounts (sameAs) |
| FAQPage | Any page with Q&A | Each question and answer; the easiest thing for AI to cite directly |
| Article | Every article | Title, author, publish date, modified date |
Why AI needs it: Without schema, AI has to guess which part is the company name and which is the price; with schema, it does not need to guess. FAQPage in particular, with its 1 question, one answer format, matches exactly how AI answers questions itself, making it the easiest to lift word for word.
How much you can do yourself: If you know a little HTML, write it while validating with Google's Rich Results Test; if you use a site builder, check whether it has a schema settings panel. Half a day.
How Get You Found handles it: All 3 types set up, and we make sure every page's canonical points back to itself. (The Aldone team was caught earlier by a bug where every canonical pointed to the homepage, a classic case of schema that might as well not exist.)
Thing 4: One Direct Answer on Every Page
Every page, within the first screen, needs a paragraph of 60 to 80 words that stands on its own and says clearly what you do, who you help and what it costs.
Why AI needs it: Generative AI chooses its sources by asking "who makes it easiest for me to lift 1 clean answer?". "We provide quality professional services" cannot be lifted; "We build 3-page websites for Hong Kong tutorial centres, HKD 6,000, with launch onboarding" can.
How to write it: Take the question customers ask most often and make it the H2, then answer it directly in the first paragraph underneath: no build-up, no adjectives. The first sentence under every H2 in this article is written exactly that way; feel free to compare.
How much you can do yourself: All of it. This needs no technical skill, only a willingness to delete adjectives. 10 minutes per page.
How Get You Found handles it: We rewrite your services and FAQ into a question-and-answer structure, confirm the facts with you, and publish.
Thing 5: Author and Date
Every piece of content should state who wrote it, when it was written and when it was updated; AI looks at these 2 things when judging credibility.
Why AI needs it: AI platforms are afraid of citing something outdated. Anonymous, undated content ranks last for credibility. Content with an author, a date and an update history tells AI it is "current" and "someone is responsible for it".
How to do it: Show the author and date at the top of the article; fill in datePublished and dateModified in the Article schema; when an old article is updated, show both dates. The author can be a team name; it does not have to be an individual.
How much you can do yourself: All of it; it is a matter of habit.
How Get You Found handles it: Every article automatically carries author, date and modified date, consistent between what is displayed and what is in the schema.
Thing 6: Consistent Business Data + sameAs
Company name, address, phone and opening hours must be identical, character for character, across your website, Google Business Profile, IG and Facebook, and linked together with sameAs.
Why AI needs it: The way AI judges "does this company actually exist" is by cross-checking several sources. If 1 says "Ltd" and 1 says "Limited", or 1 phone number has a space and 1 does not, AI lowers its confidence and would rather cite 1 competitor whose data is clean.
How to do it: Settle on 1 standard format (full company name, address format, phone format) and check each platform against it; in the Organization schema, list all your official account URLs under sameAs.
How much you can do yourself: All of it, but you need the patience to fix each platform one by one. 2 hours.
How Get You Found handles it: We align the website with Google Business Profile and social accounts, and write sameAs into the schema.
Thing 7: Ask AI Once a Month and Keep a Record
Once the first 6 things are done, ask ChatGPT, Perplexity and Google AI Mode once a month in a customer's voice, and note whether you appear and who gets cited. That is your GEO report card.
Why do it: Without a record there is no way to measure progress. AI platforms re-crawl and update on their own schedules, anywhere from a few days to a few months, so look at the trend over 3 months rather than drawing conclusions after 3 days.
How to do it: Three to 5 questions ("which company in Hong Kong does X", "how much does X cost", "how to choose X"), asked on the same day each month. Record whether you were mentioned, which URLs were cited, and which competitors appeared. 1 Google Sheet is enough.
How Get You Found handles it: At launch we hand over a 1-page baseline briefing and show you how to repeat it yourself.
The 7-Point Checklist
Here is the 1-page summary of all 7 things. We suggest printing it and ticking the boxes.
| # | Item | How much you can do yourself | Time | Done |
|---|---|---|---|---|
| 1 | robots.txt does not block GPTBot / ClaudeBot / PerplexityBot / Google-Extended | All | 5 minutes | ☐ |
| 2 | llms.txt in the root directory, first paragraph states what the company is | All | 1 hour | ☐ |
| 3 | Organization + FAQPage + Article schema, no errors in Rich Results Test | Needs a little HTML | Half a day | ☐ |
| 4 | 60-80 word direct answer on the first screen of every page | All | 10 minutes per page | ☐ |
| 5 | Author + publish date + modified date on every piece of content | All | Habit | ☐ |
| 6 | Company name / address / phone consistent across all platforms, sameAs set | All | 2 hours | ☐ |
| 7 | Monthly AI citation log (3-5 questions × 3 platforms) | All | 20 minutes a month | ☐ |
If you would rather have someone do all 7 in one go instead of ticking boxes yourself, the Get You Found launch onboarding within WKH's HKD 6,000 plan does exactly these 7 things, from llms.txt to schema, set up in one pass. See What Is a WKH GEO Website for details, or first brush up on SEO fundamentals with our earlier article 4 Strategies to Improve Your Google Search Ranking, then head to WKH Web Design Plans and Pricing.
Frequently Asked Questions
What is llms.txt? Is it essential?
llms.txt is a plain text file in your website's root directory that lists, in plain language, what your company is and which pages and articles matter, so AI crawlers can understand you quickly. It is not essential: Google has stated it does not use it, but crawlers on the ChatGPT and Perplexity side have been recorded reading it. The cost of making 1 is close to zero, so WKH does it, without claiming it is a cure-all.
Can I add schema structured data myself?
Yes, but you need a little HTML. The 3 most common types are Organization (company data), FAQPage (Q&A) and Article (articles). Google's Rich Results Test validates it instantly. If you use a site builder, some platforms have a schema settings panel; if not, you will need someone to add it to the head for you.
What happens if robots.txt blocks AI bots?
ChatGPT, Perplexity and other AI will be completely unable to read your site, meaning there is no way you can appear in AI answers. Many sites have unintentionally Disallowed GPTBot, ClaudeBot and PerplexityBot because they used an old "anti-scraping" template. Checking is simple: open yourdomain.com/robots.txt and look for those names.
If I do all 7, will AI definitely cite me?
No. The 7 things prepare the conditions for being cited, taking you from "impossible" to "possible". AI answers can differ every time, and nobody can guarantee a citation or a ranking. The most practical thing after doing them is to ask AI once a month in a customer's voice, record whether you appear, and look at the trend over 3 months.
Which of the 7 does WKH's Get You Found handle?
All 7. During launch onboarding, Get You Found sets up robots, llms.txt, the 3 schema types, the answer paragraph on every page, author and dates, and business data alignment, then hands over a 1-page "Is AI mentioning you right now?" baseline briefing. This is part of the HKD 6,000 plan, at no extra charge.