Citations: which listings count and which are ballast
Judge a listing by one question: would a customer open this page to choose a business? If yes it is evidence, if no it is ballast, and ballast still carries the risk of contradicting your details. Consumers consult an average of six review sites, so six is the budget you are allocating.
The one test that sorts them
Forget domain metrics. Ask whether a person choosing a plumber would open that page. Most listing sites fail immediately, because they were built to collect entries rather than to help anyone choose, and they read that way to a human and to a machine alike.
The pages that pass have something in common: they compare. Seven roofers in this town, in an order, with a reason for each. That is a page a person reads to the end and a page a machine can compress without losing the answer.
What the reading habits actually show
Consumers consult an average of six different review sites before deciding. The most used are Google at 71 percent, ChatGPT and other AI tools at 45 percent, and Apple Maps at 27 percent, measured across 1,002 US adult consumers in 2026.
Two things follow. First, six is the realistic budget, so a seventh entry competes with the six that already work. Second, ChatGPT at 45 percent has arrived in the same list as the map services, up from 6 percent in 2025, which means the pages an assistant reads now sit alongside the pages a person reads. A listing that helps neither is not neutral, it is a liability, because it is one more copy of your name and address waiting to go stale.
| Type of page | Would a customer open it? | Verdict |
|---|---|---|
| Curated comparison of seven businesses | Yes, to choose | Evidence |
| Google Business Profile | Yes, 71 percent use it | Evidence |
| Sector body member list | Sometimes, to verify | Evidence, if the trade uses it |
| Mass listing site with 40,000 entries | No | Ballast |
| Aggregator that copied another listing | No | Ballast, and a drift risk |
What to do with the ballast
Sort your existing entries into three piles. Keep and correct the ones a customer would open. Correct or delete the ones carrying wrong details, and prefer deletion where an old address is stuck. Ignore the harmless ones that are simply invisible, since removing them costs time and changes nothing.
The temptation is to measure by count, because count is easy. Google's own documentation is a useful corrective: structured data does not guarantee that features consuming it will appear, and by the same logic a listing does not guarantee anything by existing. It has to be read.
Questions and answers
Do low quality listings hurt me?
Is a paid entry better than a free one?
How do I check whether a directory is read?
What about listings that only exist for links?
Sources
- BrightLocal, Local Consumer Review Survey 2026 1,002 US adult consumers; six review sites on average and platform shares
- Google Search Central, Local business structured data Required fields and the statement that markup guarantees nothing
