Availability heuristic
People judge how likely something is by how easily an example comes to mind.
When estimating frequency or probability, people substitute the ease of recalling instances for the actual base rate. Vivid, recent, or emotionally charged events are easier to retrieve, so they get over-weighted, while common but unremarkable events get under-weighted. The bias is in the retrieval, not in the reasoning that follows it.
How it shows up in software
Whatever your interface surfaces most often becomes the user's model of what your product is for and how often things go wrong. A support inbox makes a team believe the loudest bug is the biggest one. A feed of recent incidents makes an on-call engineer over-index on last week's failure mode. Search suggestion lists shape what users believe is searchable.
Using it well
- Pair any anecdote in a product review with the frequency number, so the room argues about the rate rather than about the story.
- Rank empty-state suggestions by what users actually do next, not by what the team finds interesting.
- In alerting and status surfaces, show the base rate alongside the incident so a rare failure does not read as a systemic one.
- Rotate or randomise sampled examples in feedback tools, so the team does not form a model from a biased slice of recency.
Where it turns manipulative
- Manufacturing salience to sell. Repeating a rare risk in security marketing until it feels routine is fear-based selling, whatever the product.
- Surfacing a stream of recent purchase notifications to imply demand that the raw numbers would not support.
- Choosing testimonials for vividness and presenting them as typical without saying what typical is.
Where you have seen it
Google Search
Autocomplete surfaces frequent queries as you type, which then shapes what people believe others are searching for.
Stripe Dashboard
Dispute and failure views show counts and rates alongside individual events rather than a raw event feed alone.
Nextdoor
Neighbourhood posts are shown by recency and engagement, so incident posts cluster at the top of the feed.
What the research says
- Tversky and Kahneman, 1973Well evidenced
Asked whether more English words start with K or have K in third position, most participants said the first, although the third-position words are more numerous. Words by first letter are simply easier to generate.
- Lichtenstein, Slovic, Fischhoff, Layman and Combs, 1978Well evidenced
People overestimated the frequency of dramatic causes of death such as tornadoes and homicide and underestimated quiet ones such as stroke and diabetes, tracking newspaper coverage more closely than mortality statistics.
- Schwarz et al., 1991Contested
Participants asked to recall 12 assertive behaviours rated themselves as less assertive than those asked for 6, suggesting the felt difficulty of retrieval matters, not just the content retrieved.
This ease-of-retrieval variant is the shakiest part of the literature. It sits next to the social priming work that has replicated poorly, and later multi-lab attempts have found smaller and less consistent effects. The core frequency-estimation result is much sturdier than this one.
Grades are a judgement about the evidence, not about how useful the idea is. Plenty of contested effects are still worth knowing, as long as you do not cite them as settled.