AI search quick wins: What to fix on your website first

Why isn’t our content showing up in AI search?

It’s one of the most common calls we hear about. The CEO asks ChatGPT about your category, your brand doesn’t come up, and suddenly marketing and digital are scrambling to fix it.

Usually, it’s not because your team has done a bad job. It’s because AI reads your website differently from the way people do.

• How people read a page: We take in a page visually: we judge credibility from the design and can hunt down a fact even when it’s tucked somewhere odd.
• How AI reads a page: AI systems rely on what they can retrieve and interpret, so buried answers, content hidden behind scripts and conflicting information can make that harder.

How an AI system might read the Avion homepage: headings and text in order, with none of the design. Real wording and heading order; what’s extracted varies by crawler and model.

There’s good news and not-so-good news.

• The good news is that fixes can work fast. Climbing Google’s rankings can take weeks or months, but we’ve seen clients show up for specific prompts within 24 hours of a change, although it varies by platform.
• The not-so-good news is that the bigger the organisation, the harder it is to get there, because your content has to serve many audiences, stakeholders and compliance requirements at once.

You don’t need to rebuild your site, publish more content or fire your SEO team. The quick wins below focus on making what you already have clearer, more consistent and easier for machines to read.

Before you start: two common AI search mistakes to avoid

One of the most compelling questions I’ve been asked recently came from a Head of Marketing at a national insurance brand: “What’s a mistake you made, and what pitfalls are there to avoid?”

Mistake 1) Fixing before you agree what’s true

Once you know what to change, things can move fast. But in enterprise, the biggest hold-up usually isn’t the change itself. It’s discovering how much of your own information is wrong or contradictory. A fee is updated on the product page but not in the PDF. The help centre describes an eligibility rule one way and an old media release another. A comparison site still lists a product you renamed two years ago. I recently looked at a national insurer’s contact page that said the helpline was open 9am to 5pm, while the FAQ directly underneath said 8am to 6pm.

When that happens, AI either repeats the wrong fact or can’t work out what’s right and cites no one at all. And the results you promised the CEO within weeks blow out to months, because everyone is still arguing about what’s accurate.

Mistake 2) Trying to fix everything at once

AI answers informational questions most often, which is why help and support content is often the most cited part of a website. But leadership teams usually care about growth, and ‘help’ content rarely wins a new customer. Comparison is where revenue is won or lost: among Australians who use AI for money matters, 45% use it to compare financial products, according to AustralianSuper’s 2026 research. So pick the stream that matters most to your organisation right now.

Many organisations need both eventually, but starting with one makes the first 60 days far easier to plan and report on. Either way, the audit comes first, and your choice tells you which prompts and pages to look at closely.

Not sure where to start? Try a 7-step AI visibility audit

An AI search audit can take an afternoon or several months. We know how easily it blows out, so we start small: one product, the questions customers actually ask, and the facts behind them.

Step 1. Pick one product or service that matters to growth

Ideally your most profitable, or the one leadership is most worried about.

Step 2. Talk to the people who hear from customers

Ask your sales, contact centre and complaints teams for the most common queries and complaints about that product. Add whatever feedback you can get easily, such as NPS comments, Google reviews and the 30 most recent reviews on ProductReview.com.au.

Step 3. List the 20 facts customers ask about most

Usually fees, rates, eligibility, what’s covered, waiting periods, opening hours, and how to apply, claim or cancel. Cover three kinds of questions: 

• what something is
• how it compares with the alternatives, and 
• whether it’s the right choice.

Step 4. Decide what’s true, then find everywhere it’s published

Confirm each fact with product and compliance, then hunt down every place it appears: old product and campaign pages, PDFs, app store listings and comparison site profiles. Search your site and Google (for example, site:yourdomain.com.au “waiting period”), and check the sources AI cites, which often point to pages you’d forgotten existed.

Step 5. Check what AI tells customers today

Do this before you change anything. Put your questions to ChatGPT, Google’s AI Mode and the other engines your customers use, and save each answer with the question, engine and date. Record what’s wrong, how serious it is, whether it comes from your content or a third party, and who appears instead. This is what you’ll measure your quick wins against, and it’s the evidence that gets leadership’s attention.

Step 6. Log every conflict, with an owner 

Product and compliance confirm the right answer, content fixes the wording, and web or SEO retires or redirects old pages. Keep a change log of what changed, where and when. Your CMS may already keep version history, but a spreadsheet does the job for a one-product pilot.

Step 7. Find the fixes that can scale

Many large sites, especially those on a headless or component-based CMS, publish one block of content, such as a fee table or contact details, across many pages. That cuts both ways. One fix can update every page at once, which makes these blocks your fastest wins. Just be warned that this also means one error can spread just as far. If someone updates something in the block, every other place this information is referenced also needs updating. That’s how a contact page ends up with two sets of opening hours. 

Why start small with an AI visibility audit?

Starting small has benefits beyond AI search. By focusing on one high-value area, you can resolve the inconsistencies that have quietly been chipping away at customer confidence, strengthen your SEO foundations and build a measurable case for investing in GEO. It’s a manageable place to start, with the potential to prove the value of better content well beyond AI search.

Quick wins in your control: AI search content fixes

These are the fixes marketing, brand, digital and content teams can lead themselves, with product and compliance helping to confirm the facts.

Fix conflicting facts, and stop them coming back

Correct or retire every conflicting fact your audit found, merge duplicate pages, and agree one source of truth for each key fact, with an owner and a review date. Without that, the inconsistencies creep back within months.

Customers are checking, too. In WPP Media’s 2026 research, two-thirds of Australians who use AI to help make purchase decisions had asked it whether a brand’s advertising claim was true. In regulated industries, an inconsistent claim is also a compliance risk.

Build each page around the questions behind one decision

It’s tempting to treat a website like a print catalogue. But people arrive at a page with a cluster of questions about one decision: is this right for me, what does it cost, what’s covered, how do I get a quote? AI matches each question to the page that answers it most clearly.

That doesn’t mean one page per question. An income protection page can answer a dozen related questions, as long as they belong to the same decision. Give each one a descriptive subheading in the customer’s words, so AI can lift the answers one at a time. What muddies a page is mixing decisions: a promo banner competing with the headline, calls to action pointing different ways, or cross-sell for another product halfway down.

Lead with the answer people came for

What goes at the top depends on the question someone has before they land. On your homepage or About page, it’s usually what you do and who you do it for. On a product page, it’s what the product is, the key facts and who it suits, followed by cost, trade-offs and how to apply. Under every subheading, answer the question in the first sentence or two, in the customer’s words. If a section is headed with a clever campaign line, AI has to guess what it’s about, and so do your customers.

That doesn’t mean losing your brand voice. Keep headings and first sentences plain, then let your voice come through in the examples you choose, how you explain trade-offs and the reassurance you give when the honest answer is “it depends”. What doesn’t survive is the scene-setting paragraph that delays the answer.

Here’s an example for one of the questions on an income protection page: 

How long is the waiting period?

Instead of this:

At [your company], we understand that life can change in an instant. That’s why our flexible income protection cover is designed to give you peace of mind when you need it most. Waiting periods may vary depending on the options you choose and your individual circumstances. Please refer to the PDS for full details.

Try this:

You can choose a waiting period of 30, 60 or 90 days when you apply. This is how long you wait after you stop working before your payments start. A shorter waiting period means payments start sooner, but your premium will be higher.

The first version never answers the question. The second answers it in the first sentence, explains it in the second and gives the trade-off in the third, so the details from the PDS can follow underneath.

Make every section easy for AI to lift

AI pulls answers out of pages in chunks, so each chunk has to stand on its own. Use descriptive subheadings, short paragraphs, tables for numbers and comparisons, and a summary of the key facts near the top.

Don’t leave key facts only in PDFs

It’s a myth that AI can’t read PDFs. It can, we’ve seen it cite product disclosure statements. But the problem is that PDFs are long-form, so the answer to a customer’s question sits deep in dense paragraphs rather than in chunks AI can lift. Superseded versions can also continue being cited long after the web page is updated. 

Keep the PDF, because compliance usually requires it, but publish a plain-language summary of the key facts as page text and link to the PDF as the authoritative document.

Show who wrote and reviewed each page

Add a named author or subject-matter expert, a reviewer where relevant (often a clinician, actuary or compliance reviewer), and a last-reviewed date. AI systems and the people reading their answers both look for evidence that a page is current and accountable.

A small change to a page’s metadata and byline can make a big difference to how accountable it looks. Here’s an illustrative example for a home insurance page.

How to get compliance sign-off faster

In regulated industries, quick content wins usually stall in approvals rather than writing. Three things help:

• Bring compliance into the audit, so they help agree the source of truth rather than review it after the fact.
• Agree pre-approved wording for your key facts. Once the 20 facts that matter most have approved wording, the same sentence can be reused across pages without a fresh review each time.
• Show the compliance team what AI is currently saying. Inaccurate AI answers about your fees, eligibility or exclusions are a risk they’ll want fixed too, which makes them an ally rather than a gate.

Your AI search quick wins checklist

In your control

• Run a content audit and save your AI benchmark
• Fix conflicting facts
• Build pages around one decision
• Lead with the answer
• Make sections easy to lift
• Put key facts in page text, not just PDFs
• Add authors and review dates

Raise with your tech team

• Fix broken links, redirects and duplicate pages
• Keep structured data accurate
• Check AI crawlers can read your pages
• Make sure you aren’t blocking AI search bots

How to benchmark your AI search quick wins

Benchmark before you change anything. The answers you saved in step 5 of the audit are your starting point. Four to six weeks after your fixes go live, ask the same questions in the same way and compare. Look for fewer inaccurate answers, more mentions or citations, and your own pages appearing as sources where third parties used to. Your change log tells you which fixes went live when.

AI answers vary from run to run, so look for a pattern across several questions rather than reading too much into one result. To track progress over time without checking by hand, an AI visibility platform automates the work; Getting set up, also in this series, covers how to choose one.

Bonus: Four technical fixes to pass on to your tech team

These fixes sit with your web, development and SEO teams or partners. You don’t need to make them yourself, but you do need to know what to ask for, how to check where you stand and what to expect.

1) Fix broken links, redirects and duplicate pages

This is the easiest technical quick win. Research from Vercel and MERJ found ChatGPT’s and Claude’s crawlers spent around a third of their fetches on pages that returned errors, and old pages that are still live keep feeding AI out-of-date facts.

• How to check. Run a crawl with a tool such as Screaming Frog, which is what we use. It’s free for up to 500 URLs and £199 a year beyond that, at the time of writing. Or ask your website manager to export a list of every page from your CMS.
• What to expect. An enterprise site will produce thousands of URLs, because every image, video and PDF gets one too. Filter to HTML pages only, then to those that return an error (404), redirect (301) or duplicate another page, and fix those first.
• What fixing looks like. Remove dead pages properly so they can’t be indexed, rather than just taking them out of the navigation. Where a page has moved, redirect it to the most relevant live page, not the homepage.

2) Keep your structured data accurate

Structured data tells machines what a page is and who published it. It’s useful, but don’t expect it to do the heavy lifting: AI quotes the visible content on the page, not the code behind it.

3) Check that AI crawlers can read your pages

This is the fix that surprises people most. When it comes up with marketing leaders, the reaction is almost always: “I had no idea. I need to talk to my team about that.” 

At the time of writing (November 2026), the crawlers OpenAI, Anthropic and Perplexity use to fetch pages don’t run JavaScript, the code behind features like rate calculators, eligibility checkers and tabs that open on click. On some sites, the words themselves only appear once that code runs, so these crawlers see a blank or half-empty page. Google’s AI features are the exception. It’s the fix that surprises marketing leaders most, and in regulated industries it often hides the content customers care about: rates, fees and eligibility rules.

• How to check. Open a priority page in Chrome, turn JavaScript off (Settings, then Privacy and security, then Site settings, then JavaScript) and reload. Whatever disappears is what most AI crawlers can’t see. If you appear in Google’s AI Overviews but not in ChatGPT or Claude, check this first.
• What to expect. The usual fix is server-side rendering, which puts the words on the page before any code runs. It rarely means a rebuild, and nothing changes for customers or your design team: calculators and tabs still work.

4) Make sure you aren’t blocking AI search bots

Many organisations blocked AI crawlers in 2023 and 2024 over training data concerns, and some of those rules now also block the bots that fetch pages to answer questions and cite sources. Some CDNs and security tools now block AI crawlers by default, so you may be blocking them without anyone having decided to.

• How to check. Type yourdomain.com.au/robots.txt into a browser and look for names such as GPTBot, OAI-SearchBot, ClaudeBot, Claude-SearchBot and PerplexityBot next to the word Disallow. Blocks in your CDN or firewall won’t show up there, so ask your web or security team to confirm those too.
• What to expect. The change takes minutes; agreement takes longer. Ask your web, security and legal teams to decide together, bot by bot, rather than with a blanket rule. If you get pushback, it usually comes down to three concerns, and each has an answer. Most AI companies run separate bots for training and search, so you can block training and still allow search. OpenAI, Anthropic and Perplexity publish their bots’ IP addresses, so your security team can tell real bots from fakes. And rate limiting controls how often bots visit without shutting them out.

Frequently asked questions about AI search quick wins

What’s the difference between FAQ schema and FAQ rich results? 

FAQ schema is code behind a page that labels its questions and answers for machines. FAQ rich results were what Google sometimes showed because of that code: an expandable list of questions under your link in search results. Google stopped showing them in May 2026, but the schema itself is still valid.

Should we remove our FAQ schema now Google has stopped showing FAQ rich results? 

There’s no need. It does no harm, and Google and other systems may still use it to understand your page. Just don’t rely on it as a way to get cited; the visible answer on the page matters far more.

Will these changes hurt our Google rankings? 

No. Readable pages, clear structure and clean redirects are long-standing SEO good practices, and your SEO team will most likely welcome the support.

Related reading: Why AI systems often struggle to interpret large websites