Field Report 001 / AI search visibility
Published 15 June 2026
The execution gap.
Everyone in this category will audit your AI visibility and hand you a list. We went and did the work, then checked it landed. This is what thousands of shipped fixes actually look like, and where AI citations are showing up so far.
The short answer
The AI visibility category measures the problem. Almost nobody closes it. We ship the fixes, then re-fetch the live page to confirm each one rendered. That produces a checkable record of the work, and a matching record of which clients then get cited by AI. This report is the first cut of both.
3,500+
Fixes approved and shipped to client sites
Live count from our change log, kept current automatically
3,000+ (87%)
Of those shipped fixes, re-checked live on the page
We fetch the page and confirm the change actually rendered
9
Australian businesses in this sample
515 distinct pages
88%
Of the work was found by the agent, not asked for
3,225 from automated analysis
~35h
Average time from approval to live on the page
under two days
The two counts above are live from our change log. Every other figure in this report is static. From our live change log and AI visibility tracking, 26 March to 14 June 2026, pulled 15 June 2026.
01 / What the work is
What does optimising for AI answers actually involve?
There is no single switch. Being the answer comes from hundreds of modest changes that make a page easier for a machine to read, quote and trust. Across 515 pages the shipped fixes cluster into six kinds of work.
1,094
Meta and social signals
Page titles, descriptions, social cards and canonicals. The summary line an engine reads first.
696
Media and performance
Image alt text, lazy loading and dimensions. Faster, more describable pages.
694
Quotable answers
Short answer blocks and FAQ markup, written so a machine can lift the answer cleanly.
670
Structured identity
Organisation, entity, breadcrumb and definition schema. Telling AI exactly who you are.
423
AI access
llms.txt files, AI crawler permissions and sitemaps. Opening the door to the engines.
85
Content and links
Headings, inserted content blocks and internal links. The hardest to automate, smallest in volume.
The eleven busiest change types
Count of fixes approved and shipped to live client pages, by type.
Notable for a young discipline: the second most common fix we ship is the answer block, the short, machine quotable paragraph written for AI assistants. A year ago that change type did not exist in our system.
02 / What is actually getting cited
Are AI engines really citing businesses yet?
We ran 1,650 citation checks for client queries across five AI engines. 72 came back as a real citation, across five of the businesses. That is roughly one in twenty three. Most checks still return nothing, and that is the honest state of play in mid 2026. The useful part is where the wins cluster.
Citation rate by engine
Share of checks that returned a real citation for a client. 330 checks per engine.
Two engines do almost all of the work. Google AI Overviews cited a client in roughly one check in eight and reached five of the businesses, the widest spread by far. Perplexity came next. Both sit on top of a live web index and reward the same things classic search does: clear answers, clean schema and pages a crawler can actually read. The chat assistants, ChatGPT, Gemini and Claude, cited far less often and each reached only a single client.
The early, honest read: if you want to be the answer in Australia today, Google AI Overviews and Perplexity are where the effort pays back first, and they reward the structured, machine readable work in section 01. We treat that as a finding to keep testing, not a law. Five clients over a few weeks is a small base, and AI engines change how they behave without notice.
The point of all of it
We do not wait for the industry to publish a study. We run the experiment on live client sites and read our own results.
The work in this report is not theory and it is not for show. Here is what it did for real Australian businesses. Names withheld, but every figure traces to their own analytics or our change log.
A Melbourne law firm
+578%
Organic search sessions, 4,056 to 27,510 over three years.
Source: GA4, quarterly totals
The same firm
~8x
Google clicks, 839 to 6,898. Average position lifted from 46 to 18.
Source: Google Search Console
A national franchise
83 to 93
Brain Buddy audit score, and now cited in Google AI Overviews for tracked queries.
Source: Brain Buddy audit and citation tracking
A specialist medical practice moved 81 to 87 on the same audit score. We are not naming these businesses here, but every number above comes from their own Google Analytics, their Search Console, or our live change log. No estimates, no composites, no borrowed case studies.
03 / What can be proven
What can you actually prove shipped?
Across the shipped fixes the overall verification rate is high, but the average hides the real finding: verification is near total for machine readable changes and drops sharply for changes that live in page content.
Verifies almost every time
Structured, machine readable
- AI crawler access100%
- llms.txt99%
- Schema markup97%
- Meta title96%
- Entity schema95%
Harder to confirm by machine
Lives in page content and layout
- Internal link28%
- Heading update33%
- Content block insert44%
- Image dimensions55%
- Image lazy loading61%
This is not a quality gap, it is a measurement gap, and naming it honestly matters. A title tag or a block of schema either exists in the page source or it does not, so a machine can confirm it in seconds. A new internal link or a reworded heading often lives inside a theme or page builder where automated checks are less certain, so we confirm those by eye instead. Anyone claiming that every AI visibility change is automatically verified has not tried to verify the messy ones.
04 / Why it is never finished
Is being the answer a project with an end date?
No. Alongside the fixes shipped in this window, 2,982 earlier changes were superseded: replaced when the agent found a better version, or when the page, the business or the engines moved on. The work compounds and it also decays.
A meta description written in March is improved in May. An answer block that fit the old service list is rewritten when the services change. This churn is the honest reason AI visibility is sold as an ongoing service and not a one off cleanup. The page is never done because the questions people ask, and the way engines answer them, keep changing.
Fixes applied to live pages, by week
3,285 applied changes across the window. Volume scales as more pages and clients come on.
A report tells you where you are invisible. A change log proves somebody did something about it.
05 / Where the work comes from
Who finds the work, and who decides what ships?
88% of shipped fixes were surfaced by the agent's own analysis, not requested by a client or staff member. Every campaign runs a live AI agent trained on the client's business. The agent does the finding. A senior SEO does the approving. Nothing reaches a live site without that sign off.
This is the division of labour we think the category gets wrong. Sold as fully automatic, AI optimisation makes promises it cannot keep on a real client site running real plugins. Sold as fully manual, it cannot reach the volume the work actually needs. The agent reads every page and proposes; an experienced person judges and approves; the plugin applies; and the live page gets re-checked. Each fix carries the name of who approved it.
06 / What this does not show
What are the honest limits?
We would rather state these ourselves than have you find them.
Small sample. Nine businesses and 515 pages over about eleven weeks. Enough to describe the work honestly, not enough to generalise to every industry or country.
Activity is proven, attribution is not. We can show what we shipped, what we confirmed live, and the citations our clients now hold. What we cannot yet prove is that a specific change caused a specific citation. The citation numbers are real but early and modest, the citation sample is five clients, and AI engines are third party systems we do not control.
Verified means rendered, not ranked. A verified fix is one we confirmed is present and rendering on the live page. It is not a claim about position, traffic or revenue.
WordPress heavy. Most sites in this sample run WordPress with Rank Math. Our applied volume and verification rates will look different on other platforms, where more changes become tasks a person implements.
One agency's data. This is our change log, not an industry survey. We are publishing it because almost nobody else in the category can, not because it is the last word.
07 / What we take from it
What did this change about how we work?
Ask any provider for their change log.
Audits are cheap to produce. A dated, verified record of fixes that actually shipped is not. If a provider cannot show you one, you are buying a report, not an outcome.
Treat verification as a feature, not a footnote.
The gap between a high verification rate and an unverifiable promise is the whole game. Insist on knowing not just what was changed, but how it was confirmed.
Budget for maintenance, not a project.
Nearly half as many changes were superseded as were shipped. The work compounds and decays. Anyone selling a one off fix is selling you a snapshot of a moving target.
Keep a human on the approval, always.
The agent reaches the volume. The senior SEO keeps it safe and on brand. Full automation and pure manual both fail real client sites. The middle is the point.
This is ongoing research
Field Report 001 is a snapshot, not a verdict. We pull these numbers from the same live system that runs the service, so they keep moving as we ship more fixes, onboard more businesses and watch the AI engines change how they answer. The sample today is small and honest about it. As it grows, some of these findings will hold and some will not, and we will say so either way.
We will keep publishing what the data shows, including the parts that do not flatter us, because a field that is this new deserves real numbers rather than confident guesses. More reports will follow. If a claim here stops being true, this is where you will see it corrected.
Common questions
The questions this report actually answers.
Do AI citations actually happen yet?
Slowly, and from zero. We ran 1,650 citation checks for client queries across five AI engines and 72 came back as a real citation, across five of the businesses, roughly one in twenty three. Most checks still return nothing, and that is the honest state of play in mid 2026. AI engines are third party systems we do not control, so we report the early movement rather than promise a citation.
Which AI engines cite Australian businesses most?
In our data, Google AI Overviews and Perplexity do almost all of the work. Google AI Overviews cited a client in roughly one check in eight and reached the most businesses, with Perplexity next. Both sit on a live web index and reward clear answers and clean schema. ChatGPT, Gemini and Claude cited far less often and each reached only a single client. Five clients over a few weeks is a small base and engines change without notice, so we treat this as a finding to keep testing.
Is this a one-off fix or ongoing?
Ongoing. Alongside the fixes we shipped in this window, 2,982 earlier changes were superseded, replaced when the agent found a better version or when the page, the business or the engines moved on. The work compounds and it also decays, which is the honest reason AI visibility is run as an ongoing service and not a one off cleanup. This is Field Report 001, the first in a series, and we will correct any claim here in public if it stops being true.
What does verified mean in this report?
Verified means we re-fetched the live page after a fix shipped and confirmed the change is present and rendering. It is not a claim about ranking, traffic or revenue. Verification is near total for machine readable changes like schema and meta tags, and lower for changes that live in page content and layout, which we confirm by eye instead.
Field Report 001 / Close the gap
See where AI is leaving you out, then watch us close the gap.
The free AI visibility audit shows you where you are invisible when a customer asks ChatGPT, Gemini or Google's AI who to call. Then you get a change log of your own. Every campaign also runs a live AI agent trained on your business, with a senior SEO approving every change.