How to rank in AI Overviews, and what it costs you in clicks

How AI Overviews choose what to cite

Start with the uncomfortable bit: AI Overviews draw heavily from pages that already rank. So the honest answer to how to rank in AI Overviews is that the first job is earning a place in the ordinary blue links, and the second job is making your page the easiest one on that list to quote. There is no third job, no submission form, and no secret markup.

Google is unusually direct about this. Its documentation on AI features in Search states there are no additional technical requirements to appear in AI Overviews. A page has to be indexed and eligible for a snippet. That is the entry ticket, and everything after it is about being the most liftable source in the pool.

Two settings quietly disqualify pages, and both were usually added by someone with good intentions.

• nosnippet, max-snippet:0 and data-nosnippet remove a page, or the tagged block inside it, from snippet use. Google documents these as its preview controls, and they apply to AI experiences too. Teams add them years earlier to discourage scrapers, then forget. If a page carries one, it is not in the running.
• Google-Extended is not the AI Overviews switch. That crawler control governs whether your content helps improve Gemini apps and Vertex AI, not whether you appear in Overviews. Blocking it will not take you out, and allowing it will not get you in.

Ten minutes with a crawler settles both of those before you spend a quarter on content.

Answer the question in the first sentence

The highest-return change is structural: put the question in the heading, and the answer in the sentence directly underneath. Roughly forty to sixty words, self-contained, no run-up. Then explain, qualify and add the nuance below it.

Most business pages do the opposite. They warm up. Three paragraphs of context, a bit of positioning, then the actual answer somewhere near the fold. A model assembling an Overview will not dig that far in any useful way, and neither will a buyer on a phone at a job site.

What a quotable answer looks like

• The heading is the question, worded the way somebody says it out loud. Not “Our pricing approach” but “How much does a commercial fit-out cost?”
• The first sentence answers it. Give the range you genuinely quote, the yes or the no, the number of weeks. A hedge is not an answer, and hedges do not get lifted.
• The passage stands on its own. Delete “as mentioned above” and “in this guide we will explore”. If a section only makes sense after the four before it, it cannot be quoted.
• Nothing interrupts. No call-to-action box, no image, no signup bar sitting between the heading and the answer.

This is the same discipline our step-by-step AEO guide covers, pointed at one specific surface. Machines are impatient readers. So, it turns out, is everyone else.

Structure the page so a machine can lift a clean answer

Past the opening answer, build the rest of the page in liftable units. Treat every section as something that could be pulled out and shown on its own, because that is exactly what happens.

1. 1One idea per paragraph, two to four sentences. Long paragraphs bury the useful sentence inside qualifiers.
2. 2Lists for anything with parts or steps. A numbered process is far easier to summarize accurately than the same process written as prose.
3. 3Comparisons in a table. Price tiers, plan differences, option A against option B. Tables are unambiguous, and ambiguity is where citations get lost.
4. 4Real numbers wherever you have them. Ranges, timeframes, dimensions, prices in AUD with GST stated plainly. Specific pages get cited over vague ones, every time.
5. 5Visible dates. Published and last updated. Freshness carries weight on anything where the answer changes from year to year.

None of that is exotic. It is the difference between writing to be read from the top and writing to be quoted from the middle.

Schema and entity consistency: the unglamorous half

Schema markup will not push a weak page into an AI Overview. What it does is stop a good page being misread. Structured data tells a machine what a thing is instead of leaving it to infer from your wording, and inference is where good pages get skipped.

1. 1Organization or LocalBusiness site-wide: registered name, trading name, address, service area, phone.
2. 2Article with author details on posts, FAQPage on real FAQ sections, Service on service pages.
3. 3Validate what you shipped rather than trusting the plugin to have done it. Broken markup is worse than none, because you stop checking.

Entity consistency is the other half, and it is duller than it sounds. Answer engines cross-reference. If your website uses one business name, your Google Business Profile uses another, LinkedIn lists three services you dropped in 2023 and an old directory still has a dead phone number, confidence drops. Low confidence means a competitor with a boring, consistent footprint gets named instead of you.

• Same name, same description, same service list across your site, Google Business Profile, LinkedIn and any directory you appear in.
• One canonical description of what you do, specific enough to be useful. “Full-service growth partner” tells a model nothing. “Commercial electrical contractor servicing Adelaide and the Adelaide Hills” tells it exactly when to name you.
• Reviews and third-party mentions. Overviews weigh what other sites say about you alongside what you say about yourself, and the second one is cheap to write.

Schema will not make a weak page quotable. It just stops a good page being misunderstood.

A citation is not a click, and that changes the scoreboard

Here is the trade you are making. Winning a spot in an AI Overview can raise your visibility on a query while lowering your traffic from it. The Pew Research Center analyzed real browsing data from US adults and found that in March 2025, people who saw an AI summary clicked through to a website on 8% of visits, against 15% when no summary appeared.

So the pattern to expect in Search Console is impressions holding or rising while clicks soften. On a click-only scoreboard that reads as failure. On a visibility scoreboard it reads as a win. Neither reading is complete by itself, which is why the measurement needs to change before the tactics do.

Data you don’t act on is just expensive decoration. If you are going to track Overview presence, decide up front what you would actually change based on it. Three things are worth watching.

• Branded search volume. People who read your name inside an Overview often search it later rather than clicking the citation, so the demand shows up under your own brand rather than the topic.
• Enquiry sources. Add a “how did you hear about us” field to your form and read the answers. “I asked ChatGPT” and “it came up in Google” turn up more each quarter, and that field costs nothing to add.
• Scheduled prompt checks. Run the twenty questions a buyer asks before hiring you, on a fixed schedule, and log whether you get named. It is the closest thing this channel has to a ranking report, and the detail sits in our explainer on answer engine optimization.

When chasing AI Overviews is the wrong move

Plenty of Australian businesses should leave this alone for a quarter or two. Not because it does not work, but because something cheaper and more certain is sitting right in front of them.

• You have nothing in the top ten. Overviews lean on pages that already rank, so an unranked page has no path in at all. Fix the fundamentals first, and our guide to how to do SEO is the cheaper starting point.
• Your buyers search transactionally. “Emergency electrician near me” at 11pm does not produce a research-shaped Overview. Your Google Business Profile and the local pack matter far more, and that is where the money should go.
• Your revenue depends on content traffic. If ad impressions on your articles pay the bills, being summarized is a cost rather than a win, and the honest conversation there is about snippet limits, not optimization.
• Someone is selling you a guarantee. Nobody controls what an Overview says, including us. A guaranteed placement is a guaranteed invoice and nothing else, and the same test applies to any AEO proposal you are handed.

If you do have rankings and the pages are simply badly structured, this is some of the cheapest work available. It is mostly editing, it improves the page for humans in the same pass, and it still pays off if the AI hype cools off next year.

That is the work our AEO service does. And if we look at your site and decide the real problem is rankings, or tracking, or an offer nobody understands, we will tell you that instead, before you spend anything on being quoted.

Frequently asked questions