AI Search Optimisation: How to Get Found in ChatGPT and AI Overviews

Something has quietly changed about how customers find businesses. They still search, but a growing share of them never see a list of ten blue links. They ask ChatGPT which agency to call. They read Google's AI summary and stop there. They ask Gemini to compare three suppliers and only click if they want to check something.

The awkward part for business owners is that none of this shows up as a crisis in your analytics. Your rankings can look fine while your traffic softens, because the answer got delivered before anyone reached your page.

So here is what is actually going on, what genuinely changes for your website, and what is worth doing about it. Some of the advice circulating on this is opportunistic nonsense, so we will be specific about what is established and what is still guesswork.

The three surfaces you are competing in

It helps to stop thinking about "AI search" as one thing. There are three distinct surfaces, and they behave differently.

AI Overviews in Google. The summary that appears above the results for a growing share of queries. It is generated from pages Google has already crawled and indexed with its normal crawler, and it cites a handful of sources with links.

Assistants. ChatGPT, Gemini, Claude and Copilot, where someone is having a conversation rather than running a query. When these tools search the live web, they fetch a small number of pages, read them, and synthesise an answer with citations.

Answer engines. Perplexity and similar, built from the ground up to answer rather than to list. Same underlying pattern: retrieve a few sources, summarise, cite.

What all three share matters more than what separates them. Each one reads actual web pages, picks a small number of sources, and rewrites what it found. Ten results have become roughly three citations. Being one of those three is the new position one.

Your existing SEO is not wasted

The most common thing we get asked is whether this means starting again. It does not, and anyone telling you otherwise is selling something.

Every one of these systems finds pages the same way search engines always have: crawling, indexing, and judging relevance and authority. A site that ranks well is a site that gets retrieved and cited. The foundations have not moved:

• Pages that load fast and render their content on the server
• A clean, crawlable structure with a real sitemap
• Genuine authority: links, mentions, a business that visibly exists
• One page per real search intent, not twelve pages fighting each other

If your technical SEO is weak, no amount of AI-specific tinkering will help you. Everything below is what you layer on top of a site that is already fundamentally sound.

What actually changes

The difference is that you are no longer optimising to be clicked. You are optimising to be quoted. That shifts a few things in practice.

Answer the question in the first two sentences

A model reading your page is looking for a passage it can lift and attribute. If your page opens with two paragraphs of positioning before it gets to the point, there is nothing to lift.

Put the direct answer immediately under the heading, then expand. This is the single highest-return change most sites can make, and it happens to be better for human readers too.

Write in retrievable chunks

Retrieval works on passages, not whole pages. That means:

• Headings that state the question the section answers, in the words a customer would actually use
• Self-contained paragraphs, so a passage still makes sense pulled out of context
• Tables and lists for anything comparative, because structured content survives summarising
• No orphan pronouns in key sentences. "It usually costs between X and Y" is useless quoted on its own. Name the thing.

Be specific enough to be worth citing

Models tend to cite pages with concrete detail: real numbers, named tools, dates, processes, trade-offs. Generic marketing copy is unquotable, because there is nothing in it that another forty pages do not also say.

This is why our pricing guide publishes actual budget bands rather than "contact us for a quote". Specificity is what makes a page the source rather than one of the crowd.

Make it unambiguous who you are

These systems reason about entities, not just keywords. They need to be confident that your business is a web development agency in Melbourne, Australia, and not something else with a similar name.

You strengthen that with boring consistency: identical business name, address and phone number everywhere, proper Organization and LocalBusiness structured data, a real About page, named authors with genuine bios, and a Google Business Profile that matches your site. If different sources disagree about your basic facts, you get treated as uncertain and skipped.

Off-site mentions now count more than links

This is the biggest genuine shift. When a model answers "best web design agency in Melbourne", it is usually not reading agency websites at all. It is reading roundups, directories, review sites, forum threads and LinkedIn.

So the work is unglamorous and mostly happens off your own website: get listed in the directories that matter in your industry, accumulate real reviews, get included in comparison articles, answer questions in the communities where your customers actually talk. Unlinked mentions carry weight here in a way they never quite did for classic rankings.

The technical bit nobody checks

Two things are worth verifying today, because both are common and both are silent.

Check that you are not blocking the crawlers you want. AI crawlers use their own user agents, and plenty of sites blocked them in a burst of caution and then forgot. If you want to appear in ChatGPT's answers, OpenAI's search crawler needs to be able to fetch your pages.

The distinction worth understanding is that the training crawlers and the search crawlers are separate. You can decline to have your content used for training while still being cited in answers, or block both, or neither. It is a legitimate business decision either way, but make it deliberately rather than by accident.

Check that your content exists without JavaScript. If your pages assemble their content in the browser, some retrievers will read an empty shell. Server-rendered HTML is not optional here. It is one of the main reasons we build in Next.js, and a good example of how website development decisions made two years ago quietly determine your AI visibility today.

What does not work

• Keyword stuffing, in any modern costume. Repeating a phrase does not make a passage more citable.
• Volume plays. Publishing forty thin AI-written articles is the fastest way to be classed as a low-quality source and stop being cited at all.
• llms.txt. A proposed file for pointing models at your content. It costs almost nothing to add, but adoption is genuinely limited, and nobody should be charging you for it or promising results from it.
• Anyone guaranteeing AI rankings. There is no ranking to buy or manipulate. Answers vary between users and between two runs of the same question.

How to tell whether it is working

Honestly, measurement here is worse than in classic SEO, and you should be sceptical of any dashboard claiming otherwise. Four things do give you a real signal:

1. Prompt testing. Write down the fifteen questions a good customer would ask, run them monthly across ChatGPT, Gemini and Perplexity, and record whether you appear and who does instead. Tedious, and unarguable.
2. Referral traffic from AI tools. In GA4, sessions from chatgpt.com, perplexity.ai and gemini.google.com are already measurable. The numbers are small, but they convert well, because someone arriving from an answer has usually finished comparing.
3. Impressions versus clicks in Search Console. Impressions holding steady while clicks fall on informational pages is the signature of answers being consumed without a visit. That is not always a loss, but you want to know it is happening.
4. Branded search volume. If people start searching your business name directly, something upstream is recommending you.

Where to start

If you only do four things, do these, in this order:

1. Check your robots rules and confirm the search crawlers above can reach you. Ten minutes.
2. Rewrite the openings of your ten most commercially important pages so the answer comes first. One afternoon.
3. Fix your entity signals: structured data, consistent contact details, a real About page, named authors, Google Business Profile matched to your site.
4. Run the prompt test and write down where you stand today, so you have a baseline to argue with in three months.

Then keep publishing content specific enough to be worth quoting. That is the long game, and it is the same long game as before.

The short version

AI search has not replaced SEO. It has raised the cost of being generic. Ten results narrowing to three citations means the middle of the pack now gets nothing, where before it got a trickle.

The businesses winning here are not the ones chasing tricks. They are the ones with a fast, server-rendered, well-structured site, unambiguous facts about who they are, a real presence in the places their industry gets discussed, and content specific enough that a model has a reason to quote them rather than a competitor.

Want to know whether AI tools are recommending you or your competitors? Get in touch and we will run the prompt test on your industry, check your crawler access and entity signals, and give you a ranked list of what is worth fixing first. You keep the findings either way. You can also read more about our SEO services and, if you are a local trade business, the fundamentals in local SEO for Melbourne trades.

Keep reading