An Australian fertility clinic's success rate is its single most regulated claim, and on some of the biggest clinic websites that claim is not text at all. It is a picture. On 8 September 2026 we tried to fetch eight Australian IVF success rate pages. Five returned a page. On two of those five, not one success percentage appears anywhere in the readable text. The numbers sit inside chart images. A screen reader cannot read them, Google cannot index them, and no AI answer engine can quote them.
- What did we actually check?
- What did the eight pages actually show?
- Why does a picture of a number fail?
- Is an unreadable claim still a regulated claim?
- Can a patient check a clinic independently?
- What does the one clinic doing it properly do?
- How do you fix a success rate page in an afternoon?
- What this check does not show
What did we actually check?
We searched the way a patient searches. Queries about IVF success rates in Australia, clinic success rates, and live birth rates per cycle. We took the pages those searches return.
That gave us eight pages. Six belong to fertility providers or fertility doctors. One is a comparison site. One is a consumer body.
We fetched all eight on 8 September 2026 with an ordinary browser user agent and no JavaScript, which is roughly how a search crawler and most AI retrieval bots see a page.
Then, for each page, we asked four questions. Does a success percentage appear in the readable text? Is there a data table? Do the chart images carry the numbers in their alt text? And does the page point the reader at the national dataset?
A bug in our own method, and we are printing it. Our first counting pass used a pattern that required a word character straight after the percent sign. A percent sign is not a word character, so the pattern matched almost nothing and two pages that state their numbers perfectly well were scored as stating none. We caught it because a result flipped between two runs. Every figure on this page comes from the corrected pass. A measurement that changes when you look at it twice is not finished.
What did the eight pages actually show?
Three of the eight pages refused the request outright and returned a 403. That is a finding in itself, and we come back to it.
| What we measured | Result |
|---|---|
| Pages we tried to fetch | 8 |
| Pages that returned content to a plain request | 5 |
| Pages that blocked the request with a 403 | 3 |
| Of the 5 that loaded: pages with no success percentage anywhere in the readable text | 2 |
| Pages with a data table of results | 1 |
| Chart images across the whole set | 13 |
| Of those 13, chart images whose alt text contains a number | 4, and all four belong to one clinic |
| Pages that link to the national dataset | 2 |
Put the two halves together. Of the eight pages a patient can find, only three serve a success rate as text that a machine can read without running JavaScript.
The two picture-only pages belong to large, well known providers. Neither is a small clinic with a hobbyist website. On one of them, the entire success rate section is a single JPG, and its alt text is the clinic's name followed by the words success rates. No number.
The other is more striking. It carries eight chart images. Five of them have the alt text Measure01, Measure02, Measure03, Measure04 and Measure05. Those are placeholder labels from a design file. They are sitting on the images that carry the clinic's headline claim.
We tested the obvious objection before we wrote any of this. Maybe the numbers are in the alt text, and the pages are fine. They are not. Across the two picture-only pages there are nine chart images and zero of them carry a percentage in alt text. Neither page contains a single HTML table. There is no text version of the number anywhere on either page.
One more detail worth keeping. On one of the two, 692 percent signs do appear in the served HTML. Every one of them is a CSS value: widths, heights, opacities. If you count percent signs in source code and call that a data check, that page looks like the best documented page in the set. It has no data on it at all.
Why does a picture of a number fail?
It fails three different audiences at once, and the three failures have nothing to do with each other.
| Who cannot read it | What happens |
|---|---|
| A patient using a screen reader | They hear the alt text. On the pages we checked that is a filename, a clinic name, or the word Measure and a number. They never learn the success rate |
| Text in an image is not indexable content. The page can rank for the phrase success rates while containing no success rate | |
| ChatGPT, Perplexity, Google AI Overviews | An answer engine quotes text it retrieved. It cannot quote a chart. Asked which Australian clinic has what success rate, it goes to whichever page states a number in words |
That last row is the commercial one. The clinic spent real money producing that data and designing that chart. Then it published it in the one format that guarantees nothing will ever cite it.
The 403 pages have the same problem from the other end. A page that refuses a plain request has decided, usually by accident through a firewall rule, that automated readers do not get in. Some AI crawlers identify themselves and can be allowed deliberately. We wrote about which ones and how in how to rank in AI search.
Is an unreadable claim still a regulated claim?
Yes. Being hard to read is not a defence, and it is not a compliance strategy.
A success rate is a claim about the outcome of a regulated health service. Under section 133 of the Health Practitioner Regulation National Law, advertising a regulated health service in a prohibited way carries a maximum of $60,000 per offence for an individual and $120,000 for a body corporate. Putting the claim in a JPG does not move it outside the rule. It only makes it harder for the clinic to prove what it said and when.
There is also a second regulator here, and it has already been down this road.
On 14 November 2016 the ACCC published findings after reviewing the websites of Australia's major IVF clinics. Commissioner Sarah Court said some clinics used clinical pregnancy rate data to compare their success rates, where that data reflected the clinic's success in creating an embryo rather than live birth rates. The ACCC also noted clinics presenting that data alongside photographs of newborn babies.
That is the distinction the whole category turns on. A clinical pregnancy is not a baby. A rate per embryo transfer is not a rate per cycle started. Change the denominator and the same clinic's number moves a long way.
The awkward part. Both terms are still in use on the pages we checked. Three of the five readable pages use the phrase clinical pregnancy. Ten years after a regulator said the wording confuses people, the wording is still there. When the number is also a picture, a patient has no practical way to tell which measure they are looking at.
Can a patient check an Australian clinic independently?
This is what makes fertility different from every other clinical vertical we write about.
UNSW Sydney's National Perinatal Epidemiology and Statistics Unit runs YourIVFSuccess, an independent, government funded website that publishes clinic level success rates. UNSW states it allows patients to search all 90 accredited IVF clinics in Australia, and that more than 90,000 people had used it since its launch in 2021.
We pulled its sitemap on 8 September 2026. It lists 97 individual clinic pages. Whatever the exact accreditation count on any given day, the practical point holds: there is a public page for effectively every clinic in the country, and it is not written by the clinic.
So a patient reading a chart on a clinic's website can open a second tab and read an independent figure for the same clinic. No other Australian healthcare vertical has that. Dentistry does not. Physiotherapy does not. Cosmetic medicine does not.
Two of the five readable pages we checked link to it. Three do not.
Linking to the independent dataset is the single strongest trust move available on a fertility website, and it is free. It says the number survives being checked. Refusing to link to it says nothing good, and the patient can find it in one search anyway.
What does the one clinic doing it properly do?
One page in the set gets it right, and it is worth copying exactly.
- Twelve distinct percentages appear in the readable text of the page, not only in the charts.
- It has four chart images, and all four carry a percentage in their alt text. It is the only page in the set where that is true.
- It names the age bands the figures apply to.
- It uses the phrase live birth as well as clinical pregnancy, so the measure is stated.
- It links to the national dataset.
That page is not doing anything clever. It is doing the ordinary thing: writing the number down. It costs nothing, it takes an afternoon, and it is the version a screen reader, Google and ChatGPT can all use.
How do you fix a success rate page in an afternoon?
Keep the charts. Charts are good. Just stop making them the only copy of the data.
| Do this | Why |
|---|---|
| Put every number in an HTML table underneath the chart | Indexable, quotable, readable aloud, and it survives the image failing to load |
| Write real alt text on every chart | Describe what the chart shows and include the headline figure. Never leave a design file label like Measure01 in production |
| Name the measure in the same sentence as the number | Live birth or clinical pregnancy, per initiated cycle or per embryo transfer. This is the exact confusion the ACCC named in 2016 |
| Name the age band and the reporting period | A rate with no age band and no date attached cannot be compared to anything |
| Link to the national dataset | The patient will check anyway. Being the one who sent them there is worth more than being the one who did not |
| Check the page returns a 200 to a plain request | Three of the eight pages we tried returned a 403. A page an answer engine cannot fetch cannot be cited by it |
If you want the general version of this for any clinic page, we set it out in schema markup for healthcare websites and in the 30 minute healthcare website audit.
What this check does not show
Eight pages is a small sample, and we are saying so rather than dressing it up.
- It is not a survey of Australian fertility clinics. It is the set of pages that came back from patient style searches on one day. A different day returns a slightly different set.
- We have not named any clinic. The point is a pattern in how the sector publishes data, not a verdict on any provider's care or its results. Every page we checked is publicly available and anyone can repeat the test.
- We did not run JavaScript. Some of the numbers may appear for a human in a browser. That is exactly the gap: what a person sees and what a crawler retrieves are not the same document, and citation happens on the second one.
- We are not saying any of these pages breaches a rule. We did not assess the accuracy of any clinic's figures. We measured whether the figures are machine readable. Those are different questions.
- A 403 is usually not deliberate. It is normally a security product doing its job a little too well. It is still worth knowing about.
The finding that survives all of that is narrow and solid. On two of five readable pages, one of the most consequential numbers in Australian healthcare is published in a format that no machine and no screen reader can read.
