How Do You Optimise for SearchGPT and Apple Intelligence?

The two platforms need opposite work. SearchGPT is now just ChatGPT search, and OpenAI runs its own crawler and its own index, so being cited there is a distinct technical job: allow the right user agent, then publish content that answers questions in liftable form. Apple Intelligence does not run a rival web index. Apple crawls with Applebot to power Siri and Spotlight and leans on partner models for the heavy reasoning, which means Siri’s web answers still largely track what already ranks in conventional search. So one platform needs a robots.txt and content change, and the other mostly needs your ordinary Google position to be good.

That distinction matters commercially, because Australian buyers are already using these surfaces. BrightLocal’s 2026 consumer research found 45% of consumers used an AI tool such as ChatGPT, Gemini or Perplexity to find a local business in the past year, up from 6% a year earlier, with 30 to 44 year olds highest at 64%. That is a channel that went from rounding error to nearly half the market in twelve months.

First, Clear Up What SearchGPT Is Now

SearchGPT was the name of OpenAI’s search prototype. It no longer exists as a separate product: the capability was folded into ChatGPT, so what people mean by “optimising for SearchGPT” is getting cited in ChatGPT’s search answers. If an agency is still selling “SearchGPT optimisation” as a distinct service line from ChatGPT visibility, they are selling you the same work twice under an old brand name.

What matters is that ChatGPT search is not a wrapper around Google. It fetches live pages, builds its own index, and cites sources inline. Those citations are the traffic, and the crawler decides whether you are eligible.

The four OpenAI user agents, and what each one controls

OpenAI documents its crawlers publicly, and they do different jobs. Blocking the wrong one removes you from citations while doing nothing about model training, which is the mistake we see most often.

User agent	What it does	Should you allow it?
OAI-SearchBot	Builds the index behind ChatGPT search. Not used for training	Yes, if you want to be cited
GPTBot	Crawls for model training	Your call. Blocking it does not affect search citations
ChatGPT-User	Fetches a page because a user asked. Not an automatic crawler	Yes. This is a person requesting your page
OAI-AdsBot	Checks the safety of ad landing pages	Yes if you run ads to that page

The important line in that table is row two. Blocking GPTBot does not remove you from ChatGPT search, and allowing it does not get you cited. They are separate controls, and plenty of sites have quietly blocked OAI-SearchBot while believing they were only opting out of training.

Ours is explicit rather than relying on the wildcard, which is worth copying:

User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

One caveat OpenAI states directly: ChatGPT-User acts on a user request, so robots.txt may not apply to it the way it does to automatic crawlers. Treat robots.txt as the control for the index, not as a security boundary.

Apple Intelligence Works Nothing Like That

Apple unveiled the rebuilt Siri AI in June 2026, described as answering questions from the web on virtually any topic, with personal context and onscreen awareness, running on Apple Foundation Models on device and through Private Cloud Compute. Reporting in January 2026 said Apple had signed a multiyear deal for Google’s Gemini to underpin the new Siri. Apple’s own announcement credits its foundation models and does not detail the partner arrangement, so treat the Gemini layer as well-reported rather than confirmed by Apple.

For an Australian business owner the architecture detail matters less than the consequence: Apple is not building a competing web index for you to rank in. Apple’s crawling is documented in Apple’s own support notes, and the split is precise:

• Applebot is the crawler. It powers Siri and Spotlight search features. If you want to appear, it must be able to reach your pages.
• Applebot-Extended crawls nothing at all. It is a secondary token that only controls whether content already crawled by Applebot may be used to train Apple’s foundation models.

Apple states plainly that pages disallowing Applebot-Extended can still be included in search results. So the training opt-out is genuinely free of ranking cost, which is unusual and worth knowing.

The practical implication is the one most “optimise for Apple Intelligence” advice gets wrong. Because Siri’s web answers lean on conventional ranked results and partner models rather than a proprietary Apple index, the lever is your existing Google visibility, your Google Business Profile, and your structured data. There is no separate Apple ranking factor to chase.

The two platforms compared

	ChatGPT search (SearchGPT)	Apple Intelligence / Siri
Own web index?	Yes	No, relies on crawled data plus partner models
Crawler to allow	OAI-SearchBot	Applebot
Training opt-out token	GPTBot	Applebot-Extended
Does the opt-out cost visibility?	No	No, Apple confirms this explicitly
Main lever	Answer-first content plus crawler access	Conventional Google ranking, GBP, schema
Can you measure it?	Partly, via referrals	Barely

What Actually Wins Citations On Either Surface

The work overlaps more than the table suggests, because both systems reward the same content shape. Five things, in the order we do them:

1. Fix crawler access first. Allow OAI-SearchBot, Applebot, PerplexityBot and Googlebot explicitly. This is a ten-minute robots.txt edit and it gates everything else.
2. Answer the question in the first two sentences of the page. AI engines lift passages, not pages. A section that opens with a direct, complete answer is liftable. One that opens with “in today’s competitive market” is not. Every service page we build uses a short-answer block under a question-form heading for exactly this reason, and it moved our own AI-readiness score from 84 to 97.
3. Publish a llms.txt file. It is a plain-text map of your most citable pages. It is free, takes minutes, and we treat it as basic hygiene rather than a paid deliverable. See what llms.txt does for AI search visibility.
4. Make your entity consistent. Business name, address, phone and ABN identical across your site, your Google Business Profile and every directory. AI answers assemble facts about a business from multiple sources, and contradictions get you dropped rather than corrected.
5. Keep key pages fresh. Both systems favour recently updated, authoritative pages. A page last touched in 2023 competes badly against one updated last month.

For platform-specific depth, our guides on ChatGPT SEO strategy and Perplexity optimisation go further, and how AI search is changing SEO in Australia covers the wider shift.

Our Data: AI Referrals Show Up As A Channel Before They Show Up As Volume

We rebuilt a specialist Australian B2B advertising site code-first, with schema, llms.txt and an explicit AI crawler allow-list. Within 26 days an “AI Assistant” referral channel appeared in its analytics for the first time, alongside the conventional results below.

Metric	Before	After 26 days
Tracked enquiries	0 measurable	12
Mobile PageSpeed	~45/100	99/100
Pages appearing in Google daily	~8	16
AI assistant referral channel	absent	present

The full numbers are in the advertising business case study. Be honest about what that proves: a new channel appearing is early evidence the pages are being read and cited, not a revenue result. On a Melbourne cleaning business rebuild we allow-listed 20 named AI crawlers and shipped llms.txt at go-live, and its AI-readiness score moved 19 to 82 out of 100. Both are foundation work, and we say so rather than dressing it up as a lead source.

What You Cannot Measure Yet, And Who Pretends Otherwise

There is no Search Console for ChatGPT or Siri. No impression counts, no position data, no query list. What you can do:

• Watch referral traffic in GA4 for chatgpt.com, perplexity.ai and similar sources. It is real but understates reality, because plenty of AI answers are read without a click.
• Check your server logs for OAI-SearchBot and Applebot hits to confirm you are being crawled at all.
• Prompt-test manually. Ask ChatGPT the ten questions your buyers ask and record whether you are cited. It is a spot check, not a metric, and answers vary between sessions.

Anybody selling you a precise “AI visibility score” with a rank number attached is selling a proxy. The measurable ground truth today is crawl access, citation appearances in manual tests, and referral traffic. We would rather tell you that than invent a dashboard.

Frequently Asked Questions

Q: Does optimising for SearchGPT hurt my Google rankings?

No. The work is answer-first content, correct structured data, fast pages and consistent business facts, all of which conventional Google ranking already rewards. There is no trade-off to manage. The only genuinely AI-specific step is crawler access in robots.txt, and that has no bearing on Googlebot.

Q: Should I block GPTBot to stop AI training on my content?

That is a business decision, not an SEO one, and it is cheaper than most people think. Blocking GPTBot stops training use without removing you from ChatGPT search, because search citations run through OAI-SearchBot. Apple’s equivalent is Applebot-Extended, and Apple states that disallowing it still leaves your pages eligible for search results. Just do not block the search crawlers by accident.

Q: How do I know if ChatGPT or Siri is sending me customers?

Imperfectly. Check GA4 referrals for AI domains, check server logs for AI crawler visits, and ask new enquiries how they found you, because “I asked ChatGPT” is now a real answer people give. Expect the referral number to undercount, since many AI answers are consumed without a click.

Q: Is Apple Intelligence worth optimising for separately in Australia?

Not as its own project. Because Siri’s web answers lean on conventional ranked results and partner models rather than an Apple-only index, the same work that lifts your Google and local pack visibility covers it. The one Apple-specific action is confirming Applebot is not blocked.

Q: Do I need llms.txt for either of these platforms?

Neither OpenAI nor Apple requires it, and no engine has committed to using it as a ranking input. It costs minutes, helps AI systems locate your most useful pages, and we ship it as free hygiene on every build. Treat any provider charging a line item for llms.txt as a warning about the rest of the invoice.

Q: How long before AI citations start appearing?

On the rebuilds we have measured, first citations and AI referral channels typically appear 30 to 90 days after the foundations are in place, assuming the pages are crawlable and the content answers real questions. It tracks conventional indexing speed, because both systems need to crawl and trust the pages first.

---

Want to know whether AI search can even see your site today? Get a free competitor analysis - we will check your crawler access, your citable content and how your top 3 competitors are showing up in AI answers, and tell you plainly where you stand. Our AI SEO service covers the work if you want it done.