On this page
The organisation
Four industry member associations run from one office, sharing a small team: a national body, two sector groups within it, and a regional chapter of one of them. Together they hold an annual conference and awards, publish an annual capability directory, and promote several hundred member businesses to buyers at home and overseas. Promoting members all year round is a large part of what membership pays for.
The situation
Each association had its own website, and the conference had a fifth. Three of the sites ran on a platform that had reached the end of its life, so a rebuild was due either way. The associations' standing online was divided across the five, and each site's structured code described its association on its own, so no page told a search engine or AI system how the four related.
Buyers in the sector increasingly ask an AI assistant who to call. In July, before any building began, Southern Sky AI asked AI assistants the questions a buyer might ask and recorded the sources of each answer. Members were named, drawn from paid third-party directories and the members' own sites, and the server behind three of the association sites turned away requests that identified themselves as ChatGPT's and Claude's crawlers. Where a listing could be read, as on the regional chapter's site, the answer named a member straight from it.
What we built
The five sites became one, with an association switcher that moves between them the way a language switcher would. Each association keeps its own home page, About page, member directory, news, events, joining page and colours.
The content came across in full: 395 member listings for 327 businesses, 672 news stories and member updates, 7,222 images and 1,548 pages listed for search engines. A business that belongs to more than one association appears in each association's directory. Each of the 5,619 old web addresses forwards to its new page in one step, so links in old emails, articles and search results still work.
The AEO layer sits in the code and files machines read:
- Member listings as facts. Each listing is marked up as an organisation, naming each association the business belongs to, its website, logo, phone, email, street address, the regions it serves, the services it offers and its own description. 330 listings carry website, phone, email, logo and address together.
- One reference point per association. Each About page opens with a one-sentence description, a table of key facts and common questions, 52 in all across six pages, marked up so AI systems can lift them as answers. The relationships between the associations are written into the code, along with the shared chief executive.
- An open door. The site names 15 AI systems and invites them in, including those behind ChatGPT, Claude, Perplexity, Gemini, Apple, Meta and Amazon. A crawl of 1,525 pages as ChatGPT's crawler returned each one in full, with no structured-data errors, and pages begin arriving in about an eighth of a second from Sydney.
- Answers to likely confusions. Where an association shares its acronym with an unrelated trade show, its About page answers the question directly and the site's code records the difference, so AI assistants can tell the two apart.
- A guide and a record. An llms.txt file sets out who each association is and where its directory, news and events sit. Three of the associations now have Wikidata entries, linked from the site, so AI systems can confirm who they are.
- Measured from day one. Its own analytics, Google Search Console, Bing Webmaster Tools and server logs, with the same 13 buyer questions asked of five AI assistants before launch and after.
The outcome
Search engines and AI systems take weeks to move their records across after a move of this size, so the figures below are early. They are given at two points: three days and eleven days after launch.
The first three days
OpenAI's crawler made about 18,500 requests in the first two days and Anthropic's about 14,300, reading the site end to end.
Asked the same 13 buyer questions, ChatGPT cited the new site for 9, Perplexity for 6 and Gemini for 5. ChatGPT now cites the new site for each association where it had cited the earlier sites, explains how the four associations relate, including the regional chapter, and lists 35 members of the national body by name from its directory.
Asked who represents the sector's builders, Google's AI summary and Gemini now name the associations' sector group first and cite its page on the new site. In July the AI answer to the same question did not mention it.
The first eleven days
Copilot citations. In the first five days of Bing's AI Performance report, Copilot and its partners cited the site 76 times across 16 pages, with the commercial sector group's pages carrying 49 of them.
Google's AI features. Google Search Console recorded 166 appearances of the site in Google's AI features across 61 pages, and 2,610 appearances in ordinary Google results.
Live answers from member listings. In a 16-hour sample of the server log, ChatGPT fetched 128 pages live to answer people's questions, about three times the day-three rate, and 85% of those fetches were member listings.
The directory is what AI reads. AI systems made 11 to 15% of the page reads, and 41 to 47% of what they read was member listings and directories, up from 37% on day three.
Question-style searches. On Bing, the site ranked in the top three for two of the question-style searches buyers now type, asking who represents the sector and what the commercial sector group is.
Some answers have further to go, and AI answers vary between askings, so the same 13 questions are asked again at four and twelve weeks.
The standard applied
The site was built around what a member pays for: being found. That meant treating each listing as a record an AI system can trust, with the association's endorsement written into it, and treating each association's identity as something to keep intact inside the shared site. The measures were set before launch, against answers recorded before any building began, so the associations can see what changed in the answers their buyers receive.
