Transcript
Intro
An online store rewrote its product page titles in capital letters. Basically shouting. Against a control group, organic traffic went up 17.5 percent, and 20.4 percent on mobile. Then the testers went looking for those capitals in Google's results and found almost none. Searchers barely saw the change, and it still won.
So today we take on-page SEO one element at a time and ask the same question each time: has anyone measured it, and what happened? Small spoiler. Titles move traffic. A lot of the rest is a hobby.
A quick word on the receipts. Most results in this episode are split tests: similar pages divided into two groups, one changed, one left alone, both measured against a forecast. That cancels out seasons and Google updates. Each test is still one site, so read it as this can work, never as this will work for you.
Titles: where tests find effects
Titles first, because that is where the wins live. Brand moved to the front of the title: more than 15 percent. Updated Daily on a listings site: about 11. Cutting filler words on a travel site: 11. Live prices in the title: 10.
Now the part the listicles skip. The same travel site that won with live prices lost 7 percent when the price in the title was static. Same element, same site, opposite sign. A stale price is worse than no price. Titles cut both ways, which is exactly why you test them instead of copying someone's template.
Back to the capitals. Why would a title nobody saw win? The usual answer is click through rate, and it can't be that here. The testers' best guess is that Google's language processing reads capitals as emphasis, and they are careful to call it unproven. Across their whole history, capitalised titles came back positive half the time. Negative, never.
So what is the mechanism? Honestly, mixed. A title can move rankings, clicks or both, and a traffic test measures the sum. When someone tells you titles work because of click through rate, ask how they know. Verdict on title testing: works.
Quick gut check. Two title templates for ReelRadar's film pages. A is short: film name, year, brand. B promises the release date, the trailer and where to watch. Which one wins?
If you picked B, you might be right. You might also be the static price story. Nobody knows until the pages are split and measured, and that includes us.
Where do you start? The published wins point to a short list. Where the brand sits. A freshness word, only if the page really updates that often. Fewer filler words. And live data like prices, only if it is actually live.
How do you run a title test? You need lots of pages on one template, like every film page on ReelRadar, our fictional film site. Split them into two matched groups, change the titles on one, and compare against the forecast. Only have ten pages? Then before and after is all you get, so check the dates against seasons and Google updates before you celebrate.
Why Google rewrites your titles
Now a twist. Sometimes the title you tested is not the title Google shows. Back in 2021, Google said it uses your HTML title about 87 percent of the time. So roughly one result in eight shows something Google wrote instead.
Google published the reasons, and they read like a bug list for templates. Half empty titles, where only the site name survived. Obsolete ones carrying last year's date. Titles that promise things the page doesn't have. And micro boilerplate, where a whole group of pages share nearly the same title.
Google's own example of micro boilerplate is a TV forum where every season page has the same title. Season one, season two, who can tell. For a film site, that is a normal Tuesday. On ReelRadar the fix lives in the template: the season number goes in the title, and the page name is never allowed to be empty.
When Google does rewrite, it borrows from sources it lists publicly: your main headline, the H1, the og title, big prominent text, and the anchor text of links pointing at the page. Make those agree with your title element. If your H1 and your title disagree, you have handed Google the pen.
How do you spot rewrites? Search Console won't tell you, because it doesn't record the title Google displayed. So take your top pages, search their main queries, and compare what Google shows with your HTML title. Where they differ, find which of those four causes is behind it. Then fix the template, not the single page.
Here is what that looks like on ReelRadar. On the left, title, headline and og title all say the same thing, so Google has no reason to look elsewhere. On the right, the H1 says coming soon to a screen near you, and the og title says untitled. Google has three candidates and none of them agree. Guess which page gets its title rewritten.
First myth, and a favourite of audit tools. Go over 60 characters in a title, and Google punishes you.
Google's documentation says there's no limit on how long a title element can be. Long titles get truncated to fit the screen, and screens are measured in pixels, not characters. Truncation is a display matter, not a penalty. Put the words that matter first and let the tail get cut. Verdict: placebo.
Meta descriptions: clicks, not rankings
Second myth. Put your keywords in the meta description, and you will rank higher for them.
Google said in 2009 that it doesn't use the description meta tag in ranking, and it has never taken that back. Its current snippet guide makes no ranking claim at all. It says snippets come mainly from the page content itself. Verdict: placebo.
Here is why Google writes its own snippets so often, with real data. One FAQ page on the site of our patent and trademark firm client showed up in Google for 491 different searches in September 2026. One hand written description cannot be the best summary for 491 questions. So Google picks the sentence that fits each one.
Google's docs describe exactly that: different snippets for different searches. A 2020 study of 20,000 keywords caught Google rewriting meta descriptions 62.78 percent of the time. On ReelRadar, a release date search shows the date sentence, and a cast search shows the cast sentence. Same URL, different pitch every time.
So should you delete them? Removal tests came back mixed. Deleting them on hundreds of store location pages: inconclusive. Removing over long ones on an ecommerce site: about 4 percent more traffic, but only at 90 percent confidence, which is a shrug with a decimal point. Neither proves that removal wins.
Our approach is simple. Write descriptions for your important pages, aimed at the main query, because when Google does use one, it is your sales pitch on the results page. On big templated sites, generate them from real page data. And expect clicks, not rankings.
On ReelRadar, the description is a template filled from the film database: the film's name, its release status, then trailer, cast and where to watch for the reader's country. Every page gets a true, specific pitch, and nobody writes nine thousand of them by hand. When the release status changes, the description changes with it, so it never goes stale like that static price.
Snippet controls
Where you do have hard control is the snippet controls. nosnippet removes the text snippet. max snippet caps its length. And data nosnippet hides one part of the page, like a spoiler, while the rest stays quotable.
These matter more now than they used to. Google says the same controls limit what its AI Overviews and AI Mode show from your pages. So before you put nosnippet on a whole template, know that you are also opting it out of being quoted in AI answers. The idea that they only touch the blue link: placebo.
Images and video
Third myth. Alt text is a ranking factor, so stuff every image with keywords.
Google says alt text helps it understand an image, alongside computer vision and the page around it, which matters for image search. It also works as anchor text when the image is a link. For regular web traffic, though, two controlled tests came back flat.
The newer one, from June 2026, gave a classifieds site descriptive alt text on its listing images. Best estimate: plus 7.3 percent. Likely range: minus 4.2 to plus 19.4. A range that crosses zero is the statistician's way of saying nobody can tell. An older ecommerce test was inconclusive as well. Verdict for web rankings: placebo.
So write alt text for people who can't see the image. Describe what is actually in it, and image search gets the side benefit. Don't stuff it, which Google warns can look like spam. And please don't sell it to a client as a ranking project.
For a film site, image search is real traffic in its own right: posters, stills, cast photos. Google asks for descriptive file names, images placed near related text, and pages that are actually about the image. That is where alt text earns its keep.
Video has a rule that catches film sites out. To get into video results, Google wants a dedicated watch page, where the video is the main reason to visit. Its docs even list a movie review page with an embedded trailer as an example of what is not a watch page. That is the exact layout of most film sites, ReelRadar included. Awkward.
The fix is a separate trailer page for each film: the video at the top, its own title and description, and video structured data. The film page keeps its embed, which Google says is fine. It can still show as a text result or an image with a video badge, while the trailer page competes for the video results.
And trailers hosted on YouTube? Google says it may index the video on your page and on YouTube's own page, and both can appear. In practice, YouTube is a tough neighbour. So give your watch page what YouTube's page lacks, like release dates and where to watch, right under the player.
So the trailer page gets a short checklist. Video first, where nobody has to scroll for it. A title and description written for that video, not copied from the film page. Video structured data. And under the player, the things a searcher wants next, like dates and where to watch. That last part is what YouTube's page can't give them.
Author boxes
Last myth. Add author bylines and bios, and your rankings rise, because Google rewards expertise.
This one was tested twice on the same site. Round one: name and photo at the top, full bio at the bottom. No detectable impact, and if anything slightly negative. Round two: a smaller byline, no photo. No detectable impact again. Verdict for rankings: placebo.
Keep real author pages for real people, because readers and AI answers use them to judge a source. Trust itself comes from the content: first hand detail, facts a reader can check, and a reputation other sites confirm. We go deep on that in our episode on helpful content and E-E-A-T.
Wrap up
Scorecard time. The 60 character penalty: placebo. Keywords in the meta description, alt text as a ranking factor and bios as a ranking boost: placebo for rankings, though each has a real job somewhere else. Title testing: works, in both directions.
If your on-page time this quarter is limited, here is where we would spend it, from our own client work. Titles and title templates first, tested wherever you have the page count. Then the main content up near the top of the page. Then descriptions for the pages that matter. Bios, alt text and meta keywords as ranking projects get nothing.
The title test planning sheet is free on the episode page at irongoo.com. Pick a template, split the pages, write one variant, and decide before you start what result would make you keep it. Then let the data argue with you, not your gut.
