How often should you update your competitor set

Guide · Continuous Intelligence · 4 min read · last verified 2026-07-27

A competitor set is the list of vendors you actively track - in AI answers, search results, review sites, and deal notes - and it begins going stale the day it is written. How often to update it is therefore the wrong first question. The right first question is what the set is: an output of ongoing measurement, or an artifact of a one-time opinion. Get that right, and the cadence mostly answers itself - the set updates when the evidence changes, with a calendar floor for stretches when the evidence seems suspiciously quiet.

An output of measurement, not a workshop artifact

Most competitor sets are born in a strategy offsite: names on a whiteboard, consolidated into a spreadsheet, frozen into slides. Sets born this way have two defects that no refresh cadence can fix. They capture who the team already believed mattered - the competitors you knew about were, by definition, the only candidates - and they have no update mechanism, because workshops do not reconvene when the market moves.

The alternative is to treat membership as something a vendor earns through observed, recurring presence in the places your buyers actually look: the answers to their questions, the comparisons they read, the deals you contest. The method for surfacing vendors you were not watching - asking the questions your buyers ask and seeing who appears - is covered in How to find competitors you did not know you had. This piece assumes that discovery works, and is about the maintenance discipline around it: what triggers a change, what the calendar floor is, how pruning works, and who owns the list.

Event triggers that beat any calendar

A calendar refresh alone means the set is wrong for the whole interval between reviews. The stronger pattern is event-driven: specific observations that open a membership question the moment they occur.

• A vendor recurs across multiple buyer questions in a scan. One appearance is noise; recurrence across distinct questions is a candidacy. If the newcomer is not just present but ahead of you, the response is its own playbook - see What to do when an unknown vendor outranks you.
• The category's vocabulary shifts. Your positioning moves, or the market's language does, which changes which questions matter - and therefore which vendors appear in their answers. A set built for the old vocabulary silently mismeasures the new one.
• A funding round or launch lands in adjacent space. Treat news as a prompt to look, never as automatic membership. Funded entrants can buy visibility quickly, but the criterion stays the same: verified appearance in the questions you track, not the press release.
• A name recurs in lost-deal notes or sales calls. One anecdote is an anecdote. An anecdote plus corroborating appearance in tracked questions is a candidate.

The common structure: triggers create candidates, and candidates earn membership by appearing where you measure. That keeps the set evidence-shaped even when the prompt to look was news-shaped.

The quarterly floor

Triggers only fire if something is watching, and every monitoring setup has gaps it cannot see. That is what the floor is for: a full review at least once a quarter even when no trigger has fired, because silence can mean stability or can mean your watching has a hole, and only a deliberate review distinguishes the two.

The review has four parts: re-run the discovery method fresh, compare the current set against appearance data over the window, confirm each member still earns its slot, and rule on any candidates parked since last time. The floor's frequency should track how fast your surfaces move - AI answers in particular reshuffle often enough that stale monitoring misleads quickly, which is the argument developed in How often should you re-scan AI visibility. A quarter is a floor, not a target: in a category being actively re-formed by new entrants, a monthly review earns its cost.

The pruning rule

Sets only grow unless removal is a rule rather than an argument, because every name on the list has someone who once argued for it. The rule needs a form like this: a vendor absent from tracked questions and deal notes for consecutive reviews moves to a watch list - still recorded, no longer consuming weekly attention.

Pruning matters because tracking capacity is finite. Every passenger in the set dilutes the attention paid to the vendors that are actually moving, and a bloated set slowly turns the tracking dashboard into wallpaper - the trend work described in How to track competitor AI visibility over time only stays readable if the set stays honest. Log every removal with its reason, so that re-adding a vendor later is cheap and unembarrassing. Pruning is not a verdict that a vendor is gone; it is a statement that current evidence does not justify weekly attention.

One owner, logged changes

A set that everyone may edit and no one owns converges back into the workshop artifact - accumulating names from every meeting, shedding none. The fix is boring and effective: one named owner; additions require the evidence attached; removals require the reason; and a change log that can answer when a vendor entered tracking and why. The log matters more than it looks, because downstream systems inherit the set silently - share-of-voice trends, battlecards, win-loss tagging all assume the list is right, so a stale set corrupts them all at once, invisibly.

In Magrios, set membership is treated as a measured quantity: scans propose candidates with the appearance evidence attached, and the owner accepts, parks, or declines each one. The human owns the list; the measurement feeds it. Whatever your tooling, that division of labor is the real answer to the cadence question - the machine watches continuously, and the owner decides at triggers and at the floor.