An arrow, a matching label or a well-timed highlight can tell a learner where to look. This evidence guide focuses on signalling and visual cueing, rather than video length, segmentation or retrieval practice. The findings come from educational multimedia, including text-picture materials and animations. Using them to design a software demo is an application to test with that demo’s audience.
What signalling studies report about learning
Schneider et al. (2018) reported g+ = 0.53 for retention effect of signalling. Population/outcome: Media-learning retention tests; 139 outcome measures. Evidence location: Public abstract; 95% CI [0.42, 0.64]. Source: Educational Research Review, 23, 1–24.
Schneider et al. (2018) reported g+ = 0.33 for transfer effect of signalling. Population/outcome: Media-learning transfer tests; 70 outcome measures. Evidence location: Public abstract; 95% CI [0.22, 0.43]. Source: Educational Research Review, 23, 1–24.
The two estimates describe different outcomes in one review. They are not a promise for a particular interface or a direct comparison with the other reviews below.
Connecting words with pictures
Richter, Scheiter and Eitel (2016) reported r = 0.17 for effect of cues connecting text and pictures. Population/outcome: Transfer/comprehension; 27 studies, 45 comparisons, 2,464 participants. Evidence location: Public abstract and author repository description; 95% CI [0.11, 0.22]. Source: Educational Research Review, 17, 19–36.
This narrower review examined signals linking text and pictures. Its r-scale result is not interchangeable with g+ or d. The authors identified prior knowledge as a moderator; the broader 2018 review did not identify it as one. Different inclusion criteria matter.
Cognitive load and outcome-specific evidence
Xie et al. (2017) reported d = −0.11 for difference in self-reported cognitive load. Population/outcome: Subjective load ratings; 32 articles, 3,597 participants. Evidence location: Abstract; Results, Effects of cueing on SCL and learning outcomes; Table 2; 95% CI [−0.19, −0.02]. Source: PLOS ONE, 12(8), e0183884.
Xie et al. (2017) reported d = 0.27 for retention effect in the load-measuring subset. Population/outcome: Retention tests; 25 articles, 2,910 participants. Evidence location: Abstract and Table 2; 95% CI [0.08, 0.46]. Source: PLOS ONE, 12(8), e0183884.
Xie et al. (2017) reported d = 0.34 for transfer effect in the load-measuring subset. Population/outcome: Transfer tests; 29 articles, 3,204 participants. Evidence location: Abstract and Table 2; 95% CI [0.12, 0.56]. Source: PLOS ONE, 12(8), e0183884.
A practical cueing test for a product demo
Choose one task, such as connecting an account. Compare the same explanation with and without a cue that identifies the relevant control at the moment it is mentioned. Keep the wording and task constant. Check whether viewers can locate the control, recall the step and apply it to a new example. These are proposed evaluation questions, not outcomes measured by the papers for DemoBro.
Avoid treating every highlight as helpful. A cue should identify relevant information. Decorating every element can obscure that priority. For complementary design decisions, see our academic guide to demo-video learning, segmentation and recall and video versus interactive product demos.
How to read g+, r, d and a negative value
The symbols name different statistical scales. A result of g+ = 0.53 is not a 53% improvement; r = 0.17 is not 17% more comprehension. Preserve the estimator beside the value. Confidence intervals describe uncertainty under each review’s model, not the range of every learner’s result.
For subjective load, Xie et al. coded a negative d when the cued group reported lower load. That rating is not a brain scan or a measure of a specific cognitive-load component. Their meta-regressions describe study-level relationships; they do not establish a causal mediation pathway for an individual viewer.
Selection, verification and limitations
This is a curated guide to three peer-reviewed syntheses, not an exhaustive literature review or a new meta-analysis. Numerical locations, original years and estimator labels were checked on October 26, 2026. The original publication years remain 2016, 2017 and 2018.
Schneider et al. and Richter et al. were numerically verified from public publisher abstracts; Richter’s author-deposited repository description corroborates the included sample and r-scale result. Abstract-only verification does not establish a full risk-of-bias assessment. Xie et al. was checked against the open publisher full text, including the outcome-specific sample sizes, sign convention and Table 2.
The reviews can include the same underlying studies; their samples and effect sizes must not be added together as independent evidence. The subjective-load, retention and transfer subsets in Xie et al. also overlap. Different materials, learners, outcome tests and inclusion criteria limit generalisation. This page contains no customer analytics, private participant records, new experiment or measured SaaS conversion effect.
Download the reference sheet
- CSV: six figures, units, populations and source locations
- JSON: findings and full academic references
The files are an attributed reference list. Cite the original papers and keep their study contexts; the copy controls include authors, DOI, outcome and evidence location. The bibliography below links each DOI and the public access route used.