Get your site cited by ChatGPT and AI Overviews

The site in this guide: this one. On 29 October 2026 I ran an SEO, AEO and GEO pass on the rebuild of thestevenwade.com. The rebuild sits on a preview that tells search engines not to index it, and the domain still serves the old site, so there are no citations to report yet. What follows is what I changed, with the audit's numbers, and how I'll measure it.

01 Let the right crawlers in

AI search can only cite a page it can fetch. Many sites never decide this: robots.txt says nothing about AI agents, or a firewall setting blocks them unnoticed. Google's guidance on AI features asks you to make sure "crawling is allowed in robots.txt, and by any CDN or hosting infrastructure", so check both.

The decision that matters is search crawlers versus training crawlers. Each company runs separate agents, and you can treat them differently:

1. OpenAI. OAI-SearchBot decides what can appear in ChatGPT search, and OpenAI says sites that block it "will not be shown in ChatGPT search answers". GPTBot collects content that may train its models.
2. Anthropic. Claude-SearchBot indexes for Claude's search results, Claude-User fetches pages when someone asks Claude a question, and ClaudeBot collects content that may be used for training.
3. Perplexity. PerplexityBot surfaces and links sites in Perplexity's answers. Perplexity says it isn't used to train foundation models.
4. Google. AI Overviews and AI Mode come from Googlebot and the normal index. Google-Extended is only a robots.txt token. It controls whether your content trains Gemini models and grounds Gemini answers, and Google says it doesn't affect inclusion or ranking in Search.
5. Apple and Common Crawl. Applebot-Extended opts you out of training Apple's models without removing you from Apple's search features. CCBot builds Common Crawl's public archive, which many AI models have been trained on.

Fetchers a person triggers, such as ChatGPT-User and Perplexity-User, may not follow robots.txt at all. Both companies say so.

The rebuild's first robots.txt said nothing about any of them, and the audit tool flagged it. Now it names 11 agents and allows them all, training crawlers included. That was a deliberate call. The site is new and nearly invisible, and the guides exist to market my work, so being known to future models is worth more to me than keeping the text out of training. If these methods ever become a paid product, I'll revisit it.

If you want to stay citable but opt out of training, this is the split. The search and user agents go in one group and are allowed. The training agents go in a second group and are disallowed. Googlebot and Bingbot fall under the first rule. This site currently allows both groups.

User-agent: *
Allow: /

User-agent: OAI-SearchBot
User-agent: ChatGPT-User
User-agent: Claude-SearchBot
User-agent: Claude-User
User-agent: PerplexityBot
User-agent: Perplexity-User
Allow: /

User-agent: GPTBot
User-agent: ClaudeBot
User-agent: Google-Extended
User-agent: Applebot-Extended
User-agent: CCBot
Disallow: /

02 Put the answer first

Google says plainly that there are "no additional requirements to appear in AI Overviews or AI Mode", and no need for special markup or AI text files. The page has to be indexed and eligible for a normal snippet. So answer engine optimisation is mostly writing pages that answer a question cleanly, which AI answers reward harder.

What changes is structure. An AI answer rarely uses a whole page. It lifts the paragraph that answers the question, and Google says AI Overviews and AI Mode may use "query fan-out", running related searches across subtopics. So a page does better when each section answers one question without leaning on the paragraph before:

1. Answer the main question in the first two or three sentences. This guide opens with a short version for that reason.
2. Where buyers ask a question, use their words in the heading.
3. Make each section survive being lifted out alone. Name the subject instead of writing "as above".
4. Give specifics. A number, a date and a source are easier to quote than an adjective.

This site's old Field Manual descriptions were template text ("a practical AI field manual from Steven Wade"). They now describe what each page actually covers, so the summary a crawler reads matches the page.

03 Make it unambiguous who you are

An answer engine that recommends a consultant has to be sure which consultant it means. Use the same name, job title, location and services on your About page and in your structured data, and keep LinkedIn and your Google Business Profile in line with them.

Structured data says it in a form machines don't have to guess at. The useful schema.org types:

1. Person for you, with a stable @id, job title, location, knowsAbout topics and sameAs links to your real profiles.
2. Organization, or a more specific subtype, for the business, with the person as founder.
3. Article on every guide and post, with real dates and the author pointing at the Person's @id.
4. FAQPage for question and answer sections, built from the questions visible on the page.

Google says its AI features need no special schema, and it stopped showing FAQ rich results altogether in May 2026. It's still worth having. It states who's who in a form a machine doesn't have to guess at, and once it's in the template it keeps itself current. The one rule Google does set is that structured data must match the visible text.

04 Connect the pages, then publish a map

Internal links tell a crawler which pages matter and what they're about. Navigation links say little, because every page has the same ones. Contextual links, from an article to the service it supports and back, carry meaning.

Before the pass, links on this site came almost entirely from the navigation. The paid media page had 3 inbound links and the AI implementation page 2. I added a related files block to every article, guide, case file and service page, each link labelled with the target page's own title. Afterwards paid media had 11 inbound links and AI implementation 8. The two case files went from 1 to 4 each. No orphan pages before or after.

Then publish a map. Submit an XML sitemap to Search Console and Bing Webmaster Tools. That part is settled. llms.txt isn't. It's a proposal from Jeremy Howard, first published in September 2024: a Markdown file at your site's root listing your key pages with a line about each, for AI agents to read when someone asks a question. The proposal names no AI engine that reads it, no major AI search engine has committed to using it, and Google's guidance says you don't need "AI text files".

I published one anyway, for a narrow reason: most of this site's llms.txt is generated from each page's title and description and the price data, so it stays close to the site with little upkeep. If you'd have to keep one current by hand, skip it.

05 Measure it, knowing the gaps

The reporting is thin. What's available, and what each source can't tell you:

1. Buyer searches. Start from the searches a buyer would type to find you: your name, and your service plus your city. For this site that means "Steven Wade", "google ads consultant melbourne" and "ai consultant melbourne". Check each in ChatGPT with search on, Perplexity, Google's AI Mode and Copilot on the same day each month, more than once each, because the answers vary. Keep the answer and the sources it links. Trackers that collect answers through a model's API can differ from what a person sees in the app.
2. GA4. Since May 2026 the default channel group has an AI Assistant channel for visits from tools like ChatGPT, Gemini and Copilot, and ChatGPT adds utm_source=chatgpt.com to its links. AI Overviews and AI Mode are excluded and arrive as organic search. Google doesn't publish the full list of sources it counts, so add a custom channel group with an AI channel whose source matches the regex chatgpt\.com|perplexity\.ai|claude\.ai|copilot\.microsoft\.com|gemini\.google\.com, and move it above Referral, because GA4 puts each session in the first channel it matches.
3. Search Console. Since June 2026 it has a generative AI performance report, rolled out to all sites by 31 August. It shows impressions in AI Overviews and AI Mode by page, country, device and date, with no clicks and no queries. Clicks still sit in the Performance report under Web, and those AI impressions are counted there too, so don't add the two reports together.
4. Bing Webmaster Tools. Its AI Performance report, in public preview since February 2026, shows how often Copilot and Bing's AI summaries cite your pages, and the phrases its AI used to find them (not what the user typed). It's a sample, but it's the only first-party report that names the queries behind a citation.
5. Server logs. If your host provides them, check that OAI-SearchBot, Claude-SearchBot and PerplexityBot actually fetch your pages.

Don't confuse any of that with an audit score. This site's went from 56 to 61 out of 100 on the pass, and its findings from 74 to 38, most of the rest being the local test copy having no HTTPS. That measures the site, not whether anyone cites it.

The starting point is close to zero: DataForSEO finds no ranked keywords for the old site in Australia or the US. At switchover I'll submit the sitemap to Search Console and Bing Webmaster Tools, connect Search Console to OpenSEO, which I host myself, and set up its AI visibility tracking for "Steven Wade", "google ads consultant melbourne" and "ai consultant melbourne" to get a citation baseline. When there are results I'll add them here, citations or not.

Look past your own site

An answer engine can only repeat what it finds, and much of what it finds about a business isn't on its website. Directories, review sites, a Google Business Profile and other people's articles all describe you. No official document weighs these, but it follows from how retrieval works. Find the places that mention you and fix the ones that are wrong. For this site, the open item is a Google Business Profile as a service-area business in Melbourne.

Where AI fits

On this pass an AI agent drove the audit tools, extracted every page's titles and schema, pulled the keyword data and drafted the robots.txt, llms.txt and schema templates. That's the right job for it. I review what it drafted and decide what ships, including whether to let the training crawlers in.