Changing Link Tools Without Losing Your History

Sooner or later a subscription lapses, a budget moves, or an agency hands over. You switch link tools, and next month’s chart has a step in it that no event on your site produced. Someone asks what happened in August, and the answer — “we changed vendors” — is both true and, if you didn’t prepare, unprovable.

The step change is unavoidable. Two indexes are two crawls with two retention policies, which is why their counts differ at all: the mechanism in why two tools report different backlink counts. What is avoidable is losing the ability to say anything about the period before the switch.

What survives a vendor change, and what doesn’t

Sort every claim in your reporting into one of two piles before you migrate.

Survives: specific verified links. A named source URL, a named target, a date you confirmed it live, an anchor you read. Those are observations about the web, not about an index, so they remain true under any vendor. Referral sessions from analytics survive for the same reason — they were never link-tool data.

Does not survive: every count and every score. Referring domains, total backlinks, followed-link counts, authority scores at domain or page level, toxicity figures, lost-link tallies. Each is a property of one vendor’s index under one set of rules, and the new vendor’s version of it is a different measurement wearing the same label.

The practical implication is uncomfortable and clarifying: the durable part of your link history is the part you wrote down as enumerated links, and the fragile part is the part that looked most like data.

Before you cancel anything

Three things, in this order. The first one has a deadline.

1. Export everything from the outgoing tool. Full link-level export, referring-domain export, anchor export, and any historical series the interface will give you. Save the raw files with the vendor name and pull date in the filename, and keep them, because a chart you can no longer regenerate is a chart you can no longer defend. Note the export caps and filters in effect — the row-limit problem in sampling, export caps, and missing rows means a truncated export is a silently biased archive.

2. Run both tools for one overlap period. Even a single week of parallel access is enough. Pull the same target, same day, same filters, from both. This is the benchmarking overlap test and it is the only thing that lets you say later how the two counts related — an offset measured once, not guessed at forever.

3. Verify a sample of your headline claims. Take the twenty links you have publicly credited yourself with and confirm each one live. Those verified rows are the part of your history that carries over intact, and confirming them while you still have both tools open is cheaper than reconstructing them later.

What to do with the chart

Do not splice the series. Two vendors’ counts joined at a date produce a single line that implies one measurement, and the step in the middle will be read as an event. It isn’t one.

Three options that are honest:

Restart the series and keep the old one. New chart from the switch date; the previous chart archived and referenced, with the overlap comparison stated. Cleanest, and it makes the discontinuity a fact rather than a mystery.

Plot both series, both labelled, over the overlap period. Shows the reader the offset directly. Most convincing, and only possible if you did step 2.

Report deltas within each vendor, never levels across them. “Up 14 referring domains this quarter, [new vendor]” is comparable to the previous quarter’s delta in spirit if not in units, as long as you never subtract one vendor’s level from another’s — which yields the difference between two crawl regimes, nothing more.

Whichever you choose, annotate the switch date permanently. A dated note on the chart survives staff changes; institutional memory doesn’t.

Rebaselining the things you set targets on

Any threshold, alert, or target tied to the old vendor’s numbers is now meaningless and will fire wrongly.

• Alert thresholds need recomputing against the new vendor’s baseline noise level, from scratch. Carrying a threshold across a switch is the fastest way to generate a month of false alarms — see setting alert thresholds on link metrics.
• Authority-score targets are the worst offenders, because the two vendors’ scales are compressed differently and not convertible. A goal expressed as a score has to be restated, not translated. The reasoning is in Domain Rating versus Domain Authority.
• Anything expressed as a percentage needs its new denominator checked. A ratio computed over a different index is a different ratio even when the composition of your links hasn’t changed.

Hypothetically: the outgoing tool reported 812 referring domains and the incoming one reports 640 on the same day. That is not a loss of 172 domains. It’s two indexes, and the only defensible sentence is that both were pulled on the same date with filters recorded and the counts differ by 21%, cause unmeasured. Nothing in the report should show a decline. (Illustrative figures, not measurements.)

Take the opportunity to fix the definitions

A switch forces you to restate every metric anyway, which is the cheapest moment to name the rules you’d previously inherited by default.

• Which referring-domain boundary rule the new tool uses — the question in what counts as one referring domain.
• Whether the default view is live-only or historical, followed-only or all attributes, subdomains in or out.
• Which number is now your system of record, and where the pull is documented.

Write those down as the provenance block described in building a link report you can defend, and the next handover — to a colleague, to an agency, to your future self — starts from a stated definition rather than an archaeological dig.

The awkward conversation, prepared for

Someone will eventually ask whether the drop means the links are gone. The answer that works has three sentences: the counts come from different indexes and are not comparable; the specific links we verified are still verified and here is the list; the series restarts here and the overlap measurement is in the appendix.

That’s an answer nobody can dismantle, and it’s only available if you did the export and the overlap week before the old subscription ended. Afterwards it’s guesswork, which is the actual cost of an unprepared switch — not the missing history, but the inability to distinguish the missing history from a real decline. The same trap as any index-side move that looks like a site-side event, catalogued in when a metric moves but nothing changed.

What nobody outside the vendors knows

How the two indexes’ coverage of your particular neighbourhood compares, beyond what your own overlap test measured. Whether either vendor’s rules changed during your overlap week. How either index’s historical series was backfilled.

Which is the argument for measuring the offset yourself, once, on your own site, while you still can — a local number you can cite is worth more than any general claim about which tool is better, and it’s the only comparison that will still make sense to you in a year.