What exact-match percentage are people actually running right now, going into next year

Every client conversation I have about anchors ends with someone quoting a percentage they read somewhere, and every percentage is different. 1%. 5%. Under 10. One agency told a client of mine 15% was fine as long as the rest was branded.

Nobody can tell me where the number came from. I've been repeating 'under 5%' myself for about four years and I genuinely could not tell you what it's based on beyond the fact that everybody else says it.

So: what are you actually running, on sites that are currently ranking, right now. Not what you think is safe. The real number off your own profile, plus roughly what the site is and how competitive the terms are.

I'll start. Client site, mid-competition B2B, 140 referring domains. Exact match on the main term is 4 links, so about 3%. Partial match maybe 18%. Rest is brand, naked, and junk. It's ranking 2 and 3 on its two main terms and has been for eight months.

Depends what you're running, as ever. But I'd like to see the spread.

Five sites I have full profile access to, all currently ranking top 5 for their main commercial term. Anchors counted per referring domain, not per link, which matters and is why published figures vary so much.

Site	RDs	Exact	Partial	Brand + naked
A	96	2%	14%	84%
B	210	6%	22%	72%
C	61	11%	9%	80%
D	480	1%	8%	91%
E	133	7%	31%	62%

Range is 1% to 11% and every one of them ranks. If there is a threshold in that range, my sample does not find it.

What I do notice is that exact match percentage falls as total referring domains rise, which is either a natural consequence of bigger sites picking up more unmanaged links, or evidence that the tolerable absolute number is roughly constant and the percentage is the wrong unit entirely. Site C at 11% is 7 links. Site D at 1% is 5 links. Those are much closer to each other than the percentages suggest.

I would put money on the absolute count mattering more than the ratio, and I have no way to test that.

seven links versus five. that's the whole spread. everyone's been arguing about a rounding error for ten years

The number came from a 2013 blog post and I can more or less tell you which one.

After the first Penguin there was a wave of people reverse-engineering penalised profiles, and someone published a chart showing hit sites clustering above about 60% exact match and clean sites below about 15%. That got repeated, then rounded, then made stricter with every retelling, because being stricter is free and sounds more expert. 60 became 15 became 10 became 5 became 'under 2 to be safe'. Twelve years of Chinese whispers and not one new measurement.

What's actually true, as far as I can tell from watching this since before the number existed: profiles that got hit in 2012 were 70% exact match with the same anchor repeated. Not 12%. The failure mode was so far past anything anyone here is doing that the modern debate is about a region nobody has ever seen a problem in.

MetricsOverMagic's absolute-count theory is more interesting than anything in the percentage debate. I'd add a third variable: whether the exact matches are spread across time and source or arrived together from one place. I've seen a site with 20-odd exact matches be completely fine because they'd accumulated over six years, and that's not a shape you can express as a percentage at all.

Which blog post? If it's the origin of a decade of advice, name it.

I want to make the measurement problem explicit, because half the disagreement in threads like this is people counting different things and not realising it.

Four ways to count the same profile

Take a site with 100 referring domains and 340 total links, of which 25 links across 6 domains are exact match.

• Per link: 25/340, so 7.4%
• Per referring domain: 6/100, so 6%
• Per link, sitewide links collapsed: possibly 8/323, so 2.5%
• Per referring domain, only counting dofollow: maybe 4/71, so 5.6%

Same site, four defensible methods, 2.5% to 7.4%. That is most of the spread people argue about, before anyone has ranked anything.

What I would actually count

Per referring domain, dofollow only, sitewide links collapsed to one. That approximates the unit of endorsement rather than the unit of markup, and it is the only version where the number stays stable when a site changes its template.

On the threshold question

I think MetricsOverMagic is right that the absolute count is the more likely variable, and I would go further: the thing being assessed is almost certainly not the ratio at all but whether an individual link looks placed. A single exact-match anchor in a paragraph of otherwise natural text on a site with no other commercial outbound links is not made suspicious by what the other 99 links say. The profile-level ratio is a proxy we invented because it was the only thing the tools could show us in 2013.

I have not tested any of this and I do not know how anyone would. Flagging that clearly, because the above reads more confident than the evidence behind it.

Summary: fix your counting method before you argue about your number, and consider that the number may not be the thing.

mine is like 40% and nothing has happened lol

That's more useful than it looks, if it's true. What's the site and how many RDs?

IME the people running numbers everyone says are fatal and getting away with it are the only real evidence available in this whole area. Everything else in this thread is people who chose a conservative number reporting that nothing bad happened, which tells you nothing - you can't learn where the cliff is by staying a mile from it.

A site at 40% that's been fine for a year is worth ten of the profiles at 3%.

Recounted mine per referring domain with sitewides collapsed like hexforge said and it went from 9% to 4%. I'd been worrying about the wrong number for a year.

Agreed, and it's why my table is nearly worthless. Five sites all chosen to be safe.

Coming back to this after a month because someone asked me in PM what number to run next year and I couldn't give one, which annoyed me enough to think about why.

The reason nobody can answer is that the question assumes a rule exists and is being enforced against a threshold. Everything I've seen since 2017 says it isn't a threshold, it's a classifier, and a classifier doesn't have a number you can stay under. It has a set of things that co-occur - anchor, source, surrounding text, who else that source links to, when the links arrived. You can be at 20% and invisible because every one of those checks out. You can be at 4% and discounted because all four of your exact matches came from the same panel in the same week.

Which means the honest answer to 'what percentage is safe next year' is that the percentage was never the thing, and the people quoting you 5% are quoting a 2013 chart at a system that stopped working that way a decade ago.

I'll say the same thing again in eighteen months and someone will start a new thread asking for a percentage. That's fine. It's a comfortable question and the real one isn't.

OP back to close my end. I asked for a spread and I got 1% to 40%, all of it ranking, which is not the answer I expected and is more useful than the one I wanted.

What I've changed: I count per referring domain with sitewides collapsed now, and I've stopped quoting a percentage to clients entirely. What I say instead is that I'll take an exact match when the placement is good enough that I'd want the link anyway, and I won't buy one from a source I wouldn't buy a branded link from.

That's harder to sell than a number. It's also the only thing this thread supports.

Still think old_growth_seo overstates the classifier thing - he's describing a system nobody in here can observe, which is convenient. But I can't produce a threshold either, so I'll leave it there.