Reinstating context from encoding to retrieval increases true recognition (e.g., testing object memory on reinstated versus switched background scenes), but context reinstatement also increases false recognition of similar objects (a “context reinstatement illusion”; see Doss et al., Psychological Science, 29, 914–925, 2018). Here we report three experiments extending these context reinstatement effects to semantically associated words (i.e., the Deese/Roediger-McDermott (DRM) false memory task), thereby demonstrating that the context illusion generalizes beyond the object pictures used in prior work. We found robust context reinstatement effects on both true and false recognition, even though participants were told that the encoding and retrieval context would not always match (Experiments 1 and 3). Moreover, context reinstatement boosted claims of “recollecting” details, thereby boosting the typical DRM illusion. By contrast, when we gave a stronger warning to avoid context as misleading (Experiment 2), context reinstatement effects on both true and false recognition were smaller and unreliable, suggesting that strategic attempts to minimize this illusion during retrieval came at the cost of context reinstatement’s benefit to true recognition. Overall, our demonstration of a context reinstatement illusion using the DRM task provides another example of how context reinstatement can increase both true and false recognition. These results also provide new evidence that item-context conceptual associations can drive these misleading effects of context reinstatement during memory retrieval.