Most SEO advice gets passed along without anyone checking it. One site publishes a list of ranking factors, ten more rewrite it, and a year later the same claims sit on page one for every search about what works in SEO. Very little of it comes from watching real business websites over time.
So we checked our own. Between April and September 2026 we tracked 1,982 keywords across 62 client websites and logged every piece of SEO work those sites received in the same months: backlinks, new pages, rewrites, title changes, schema, internal links. Then we compared sites that got each kind of work with sites that didn’t, keyword by keyword, from the same starting positions.
Some results back up what most SEOs already believe. Others don’t, and a few tasks that agencies bill for every month showed no measurable effect at all. We have left every client name out. The data is first-party and, as far as we can tell, nobody has published a like-for-like comparison of this kind for small and mid-sized businesses.
The short version
- Backlinks moved rankings, but only once a site had three or more live links over about five months. Those sites had 55% of their keywords in the top 10, against 44% for sites with no new links.
- Sites that got one or two links did no better than sites that got none.
- Rewriting existing pages around what searchers want was linked to a 6 point gain. Adding more words to pages without rewriting them showed nothing.
- New service and location pages were behind the biggest organic traffic growth in the study.
- Title and meta tweaks, H1 changes and blog posts showed no measurable ranking benefit.
How we ran the study
The websites
62 business websites, most of them Australian, covering trades, professional services, finance, education, engineering and equipment hire. We dropped one site that moved to a new domain partway through, because its links pointed at one address while its rankings were tracked on another.
The rankings
1,982 tracked keywords. For 49 sites the starting point was 8 April 2026; for the other 13 it was 1 July. The end point was each keyword’s typical position across the second half of September, which irons out the daily wobble every rank tracker shows.
The work
We opened every article that was supposed to carry a link and only counted links that were live, dated to the day they went up. Content came from the publishing calendar, sorted into new pages, rewrites of existing pages and blog posts. On-page changes came from comparing monthly crawls of every site from July onward: title and meta description changes, H1 changes, pages that grew by a quarter or more, new schema, new internal links inside body copy and new URLs.
The comparison
The main measure is simple. At the end of the window, what share of a site’s keywords were in the top 10? We compared that share between sites that received a type of work and sites that didn’t, using only keywords that started in the same rank band. That last part matters. A keyword sitting at position 3 in April usually stays near the top whatever happens, so comparing like with like stops already strong sites from flattering the result.
Anything that looked like an effect then went through more checks. We re-ran each result with one site removed at a time, then with every activity controlled for at once. We ran a placebo test (explained below), checked whether the page that got the work beat the same site’s other pages, and looked at organic users in Google Analytics.
This is not a controlled experiment. Some groups were as small as seven sites. Where a result survived every check we say so, and where it only showed up in one test we call it a hint.
Finding 1: backlinks work, once you pass a threshold
Sites that picked up three or more live links during the window finished with 55% of their keywords in the top 10. Sites with no new links finished on 44%. The gap showed up at every starting rank: fewer page-one keywords slipped (17% against 23%), and more keywords from positions 11 to 50 made it onto page one.
It also held up under pressure. Removing any single site left the gap between about 10 and 17 points, and it stayed clear when we controlled for every other type of work, the site’s starting ranks and how long it had been with us.
One or two links did nothing
This surprised us. Sites with one or two new links ended with 38% of keywords in the top 10, below the no-link group, and lost 36% of their April page-one keywords. We can’t fully explain the dip. It may say more about which sites happened to get small batches than about the links. What we are confident of is that a link here and there did not register in rankings.
Three links means one every six weeks or so
The sites in the 3+ group got between three and six links over roughly five months. That works out to about one every six weeks, which is a far smaller number than most link packages sell. What the group had in common was a steady supply.
Links built in the twelve months before the window had no relationship with ranking movement during it. Our reading is that fresh links arriving matter more than a pile of old ones, though six months of data is too short to say how long a link keeps pulling its weight.
The placebo test
Every study like this has the same weak spot. Maybe stronger sites get more links and would have climbed anyway. To test that, we looked at sites whose links only went live in September, too late to affect a ranking measured in the same month. If links were just a badge worn by strong sites, this group should have matched the early-link group.
It didn’t. The September-only sites were about 3 points ahead of the no-link sites before their links arrived and about 4 points ahead at the end, which is close to flat. The sites that got their links between March and August went from almost 4 points ahead at an early September checkpoint to almost 12 points ahead by late September. Plenty of that lift showed up late, which fits the common experience that links take a couple of months to register. We would want another round of data before leaning on the timing.
The placebo group does tell us something uncomfortable. Those sites were already a few points ahead before any links arrived, so part of the headline gap is stronger sites attracting links. Taking that out, our best estimate of the real effect is closer to 8 points than 11. That is still the largest effect of anything we measured.
The linked page didn’t move. The whole site did.
We traced 34 links back to the 26 pages they pointed at across nine sites, then compared those pages with the same sites’ unlinked pages over the same weeks. If a link lifted its own target, linked pages should have ranked well above the middle of the pack. They sat at the 45th percentile, where 50 means no difference at all. Homepages that received links beat the site median in only 2 of 12 cases.
Put that next to the site-wide result and the picture is consistent. In our data, links raised the site as a whole rather than the page they landed on. If you have spent hours arguing over which URL a guest post should link to, the time is better spent getting the next link.
Traffic backed this up. Sites with three or more links grew organic users by 17% between late June and September while sites with no links fell 11%. In the five weeks after a batch of links went live, organic users ran about 7% ahead of comparable sites, and about 11% ahead when the batch had two or more links.
Finding 2: rewriting pages helped, padding them didn’t
Twelve sites had existing pages properly rewritten during the window: new copy built around what people searching for that service want to know, extra sections that answer real questions, and structure that answer engines can quote. Those sites finished with 48% of keywords in the top 10 against 42% for the rest. The 6 point gap held with any one site removed (5 to 8 points) and stayed positive, though less clear-cut, once every other activity was controlled for. We would call it likely rather than certain.
Our crawls also picked up a different kind of change: pages that simply got longer, by a quarter or more, without a rewrite. Those sites finished 1.6 points behind the sites that didn’t do it. A proper rewrite starts from the question the searcher asked, while padding starts from a word count target, and Google appears to reward the first one.
One caution. The rewritten pages did not outrank their own sibling pages by much (0.7 positions on average, better in only 3 of 9 sites). As with links, the gain showed up across the site more than on the page that got the work.
Finding 3: new pages drove the biggest traffic gains
New service and location pages made no difference to keywords we were already tracking, which is what you would expect. A new page for a suburb or an industry goes after searches the site wasn’t competing for, so it can’t show up in rankings for terms that already had a page.
Traffic shows what they did. The four sites that launched new service or location pages grew organic users by 26% while everyone else fell 6%. The clearest single case was a business that published 10 location pages and 22 industry pages. None of them had a single backlink pointing at it, and the site’s non-brand impressions in Search Console grew about eightfold.
Four sites is a small group, so treat the percentage with care. The direction is hard to argue with, though. If a business sells a service or covers an area that has no page on its website, it has no page to rank for that search.
Finding 4: internal links look promising
When five or more pages on a site gained new links inside their body copy from related pages, the site finished with 57% of keywords in the top 10 against 48% for the rest. That 8.6 point gap is based on only seven or eight sites and its likely range runs from under 1 point to almost 19, so we call it promising rather than proven. Adding internal links to one or two pages showed nothing, and we didn’t see a traffic signal yet.
Finding 5: schema didn’t move rankings, but traffic rose
Adding schema markup made no difference to rankings (0.8 points). The seven sites that added it did grow organic users by 9% while comparable sites fell 8%. That could be rich results pulling more clicks from the same positions, or those sites simply having a better season. We lean towards the first explanation because rankings stayed flat while visits went up, but seven sites can’t settle it.
What showed no measurable effect
Title and meta description tweaks
Sites where our crawls picked up targeted title and meta rewrites on three or more pages finished 5.3 points behind the rest. A full sitewide rewrite came out 5.6 points ahead, but the likely range crossed zero, so we can’t call it an effect. Titles clearly shape how many people click a result, and rank tracking can’t see that. As a monthly task meant to lift rankings, though, we found nothing.
H1 changes
Sites with changed H1s finished 6.2 points behind, and keywords on the pages that got a new H1 did slightly worse (1.4 positions) than the same site’s other keywords. We don’t think changing an H1 hurts a page. We just can’t find any sign that it helps.
Blog posts
Taken alone, sites that published blog posts looked 5.2 points ahead. Once we controlled for everything else those sites were getting, the gain disappeared and turned slightly negative. That fits what blogs are for. Most tracked keywords for service businesses are commercial (“commercial electrician Melbourne”, not “how to rewire a house”), and blog posts rarely target those searches. Blogs can still earn links and answer questions, but they aren’t a lever for ranking money keywords within six months.
Ticking off the monthly checklist
Sites where every planned monthly task was marked complete finished 6.5 points ahead, but the likely range crossed zero. Completing a checklist is a sign someone is looking after the account. It only helps rankings when the list has the right things on it.
What we would spend a small SEO budget on
If a business came to us with a modest monthly budget and asked where it should go, this data points to a fairly clear order:
- A steady flow of links from real, relevant sites. At least one every six weeks, kept up. One or two links bought in a burst looked like money wasted in our data.
- New pages for services and locations the site doesn’t cover yet. This is where traffic growth came from.
- Proper rewrites of the pages that matter. Start from what the searcher needs to know, not a word count.
- In-content links between related pages, added across several pages at once rather than one at a time.
What we would stop paying for as ranking work: monthly title tweaks, H1 swaps and padding pages with extra paragraphs. Some of those are worth doing once for click-through or readability. None of them showed up in rankings.
The limits of this data
- It is observational. We compared sites that got different work, and we didn’t randomly assign it.
- Groups were small: as few as 7 sites and rarely more than 20.
- Six months is a short window for SEO, especially for links built in July and August.
- Tracked keywords lean towards terms each site already targeted, which undersells new pages and blogs.
- Traffic is seasonal. Tutoring businesses spike in September, while water and hire businesses pick up in spring, so we treat the traffic figures as supporting evidence only.
- Many sites got several kinds of work at once, which makes the effects harder to pull apart.
We will run it again in six months, with keywords added for every new page on the day it goes live and every link logged against the page it points at.
Questions about the study
Do backlinks still work in 2026?
In our data, yes. Sites with three or more new live links over about five months had 55% of keywords in the top 10 against 44% for sites with none. After allowing for stronger sites attracting more links, the real effect looks closer to 8 points, which is still the biggest of anything we measured.
How many backlinks does a small business need each month?
Fewer than most packages sell. The sites that pulled ahead got three to six links across five months, about one every six weeks. Keeping it up mattered more than the number in any one month, and one or two links on their own showed nothing.
Do title tags and meta descriptions affect rankings?
We found no measurable ranking effect from rewriting them, either on selected pages or across a whole site. They can still change how many people click, which rank tracking doesn’t capture.
Do blog posts help a service business rank?
Not for commercial keywords within six months, in our data. The small gain for sites that blogged disappeared once we accounted for everything else those sites were getting. New service and location pages did far more for traffic.
Is this study proof of what works in SEO?
No. It is observational data from 62 websites, and correlation isn’t proof. Backlinks came through a placebo test and every other check we threw at them, and page rewrites came through most of them. Everything else here is a hint at best. It is still more than most SEO advice is based on.
