How to measure link building ROI: a framework for SEO and AI visibility

The usual link building report is a spreadsheet of URLs and DRs. It proves something was bought. It doesn't say whether the money did anything, and that's the question a finance director, a client or a founder actually asks.

Answering it properly takes some setup before the campaign starts. Most of the work is deciding what you'll measure, on which pages, against what baseline.

Why link ROI is hard to pin down

Four reasons, and they compound.

The lag is long. Ahrefs tracked 2 million pages and found only 5.7% reached the top 10 for at least one keyword within a year. Links speed that up, but you're still measuring over quarters.

Links aren't the only input. Content changes, technical fixes, competitors and Google's own updates all move the same rankings. In 2026 so far Google has run a March core update, a May core update and spam updates in March, June and August. Any of those can swamp the effect of a quarter's placements.

Some links stop counting. Google's spam updates documentation says that when its systems neutralise spammy links, the benefit they gave is lost and can't be regained. A link can be live, indexed and worthless at the same time.

And a growing share of the value never becomes a click. Pew Research found Google users clicked a traditional result in 8% of visits when an AI summary appeared, against 15% without one, and clicked a link inside the summary in about 1%. BrightEdge reported that a year after AI Overviews launched, impressions were up more than 49% while click-through was down nearly 30%. If your brand is named in the answer, you've had exposure that analytics won't record.

Leading and lagging indicators

The time ranges are our working expectations, not guarantees. Competitive head terms take longer. Long-tail pages on a site with existing authority can move within weeks.

On referring domains: Ahrefs found a clear correlation between the number of sites linking to a page and its traffic, and the same study found 96.55% of pages in its index get no Google traffic at all. Referring domains are a good leading signal. They aren't the result.

A measurement framework

Before the first order, write this down for each campaign. It takes an hour and saves a lot of arguing later.

Use control pages

This is the single most useful habit we'd recommend. Pick target pages for the campaign, and a matched set of similar pages (same template, similar traffic and rankings) that get no new links. If the target pages climb and the controls don't, you've got a reasonable case. If both climb, it was probably an update, seasonality or something else on the site. It isn't a controlled experiment, but it's far better than looking at a sitewide traffic graph and hoping.

Fix the keyword set

Choose the keywords per target page before you start and don't change them mid-campaign. Otherwise the report drifts toward whichever terms happened to go up.

Turning it into a number

The simplest honest version: incremental organic value on target pages, over a period, divided by what you spent.

A hypothetical example, to show the arithmetic. A SaaS company spends £6,000 over a quarter on placements pointing at three feature pages. Six months later those pages get 1,400 more organic sessions a month than before, and the control pages are flat. The pages convert organic visitors to trials at 2%, and 20% of trials become customers worth £900 in first-year gross margin. That's 28 trials and about 5.6 customers a month, or roughly £5,000 a month in first-year margin. On those numbers the spend pays back in under two months of the uplift. Change any assumption and the answer moves, which is why you write the assumptions down with the finance team first.

Don't give links all the credit. If the pages also got rewritten during the period, say so and split the credit, or don't claim a number at all.

When nothing moves

Say it's month five. The links are live and indexed, referring domains went up as planned, and the target pages haven't budged. What now?

Check the links first, then the pages. Are the linking pages themselves indexed and getting any traffic? A link from a page Google ignores won't do much. Did the publications drop during a spam update after you bought from them? If several did at once, that's your answer, and a reason to tighten how you vet publishers.

If the links look fine, the bottleneck is probably the page. Thin content, a weak match with search intent, or a competitor set with ten times the referring domains will all cap what links can do. Links amplify a page that deserves to rank. They rarely rescue one that doesn't, and we'd rather tell you that in month five than keep selling placements into a page that can't convert them.

Attribution caveats for 2026

AI traffic shows up in odd places. Google says traffic from AI Overviews and AI Mode is reported in Search Console's Performance report under the Web search type, mixed with everything else. Will Critchlow's May 2026 round-up for SearchPilot found AI Overview clicks are indistinguishable from regular organic in GA4, most AI Mode clicks arrive without a referrer and land as Direct, and ChatGPT's web traffic shows a chatgpt.com referrer, with links tagged utm_source=chatgpt.com but no medium, so tagged visits land in Unassigned in GA4's default grouping. ChatGPT's mobile apps mostly show as Direct.

Practical upshot: build a custom channel group for known AI referrers, watch Direct traffic to deep pages (nobody types those URLs), and don't expect the numbers to be complete.

On value per visit, there's one widely quoted figure. Semrush estimated the average AI search visitor is 4.4 times as valuable as a traditional organic visitor, based on conversion rate. It's a projection from a vendor that sells AI visibility tools, built on extrapolated data. We wouldn't put it in a business case. Measure your own.

Measuring AI share of voice without fooling yourself

AI answers are unstable. SparkToro and Gumshoe ran 12 prompts through ChatGPT, Claude and Google's AI almost 3,000 times and found less than a 1 in 100 chance of getting the same list of brands twice. Position in a list meant almost nothing. How often a brand appeared across many runs held up much better.

So run the same fixed prompt set repeatedly, per engine, and report:

• Mention rate: the % of runs where your brand appears
• Share of voice: your mentions as a share of all brand mentions in your competitor set
• Citation share: how often your site or the publications you placed on are cited as sources
• Sentiment of the mentions, read by a person on a sample
• All of it split by engine, because ChatGPT, Perplexity, Gemini and AI Overviews draw on different sources

Then link the trend back to placements. If the publications you bought from start appearing as citations for your prompts, and your mention rate rises in the engines that cite them, you have a plausible causal story. Our guide to link building for AI search covers how to choose those publications in the first place.

Reporting cadence

Monthly: spend, delivery, link health, referring domains, referral sessions, early ranking movement. Quarterly: positions against the fixed keyword set, organic clicks against controls, conversions, AI share of voice, and a decision on what to change. Agencies reporting to clients should agree this split up front; it stops the month-two "where are the results" conversation. More on that on our agencies page, and on pacing spend in link building budget and velocity.