Evidano is an AI-powered qualitative data analysis platform that ingests mixed digital content and produces reproducible, auditable syntheses for research teams. Fast, reproducible assessments of online health information are critical after the PLOS Digital Health study (Published July 9, 2026) that evaluated 144 digital and AI resources for adults with cancer who care for children. The study found Google web results (n=39) scored highest for quality (mean mDISCERN 3.74, GQS 3.72) while AI (n=34) and TikTok (n=47) scored lower (AI mDISCERN 2.77; GQS 2.32). If you run qualitative reviews of websites, videos or LLM outputs, this post shows how to operationalise an AI-enabled qualitative analysis (what to automate, how to compare segments, and how to measure readability and inclusivity) using Evidano to cut manual work and produce decision-ready reports. Note: this post is research-focused and non-diagnostic.
Key Takeaways
This post shows how to reproduce Anand et al.'s PLOS Digital Health comparison and how an AI-enabled qualitative workflow highlights quality, readability, and inclusivity gaps across websites, AI outputs, and short-form video.
The most reliable headline findings are: traditional Google web pages scored higher on quality metrics, AI outputs were shorter and lower-scoring, and estimated reading age across written materials was about 15 years.
- Anand et al. screened 690 items and included 144 resources (Google web 39; AI 34; YouTube 24; TikTok 47), study published July 9, 2026.
- Quality metrics reported: Google web mDISCERN mean 3.74 and GQS 3.72, AI mDISCERN mean 2.77 and GQS 2.32.
- Readability and length: estimated reading age approximately 15.1 years, mean reading time Google 11:45 min versus AI 02:09 min (mean words Google 3, 025 vs AI 586).
Fast take + source
Fast take: Anand et al., PLOS Digital Health (Published July 9, 2026) compared Google web, AI chat outputs, YouTube and TikTok and concluded traditional websites had higher quality and reliability, but written materials often required a reading age of about 15 years and lacked inclusive coverage.
Read the full paper: PLOS Digital Health.
- Study sample: 144 resources (Google web 39; AI 34; YouTube 24; TikTok 47).
- Key quality metrics: Google web mDISCERN 3.74 and GQS 3.72; AI mDISCERN 2.77 and GQS 2.32.
- Readability: estimated reading age about 15 years; mean read time Google 11:45 min vs AI 02:09 min.
Findings snapshot
| Metric | Value | Source / Note |
|---|---|---|
| Total resources screened/included | 690 screened → 144 included | Searches completed July–Oct 2025; Published July 9, 2026 |
| Quality: Google web (mean) | mDISCERN 3.74; GQS 3.72 | Higher reliability vs AI/TikTok (P <.05) |
| Quality: AI (mean) | mDISCERN 2.77; GQS 2.32 | Lower overall quality; shorter output length |
| Readability (estimated) | Reading age ≈15.1 years (Google & AI) | May exceed recommended public reading levels |
| Reading time (mean) | Google 11:45 mins; AI 02:09 mins | Google content much longer (mean words 3, 025 vs AI 586) |
What the study measured and why it matters
This section explains what Anand et al. measured and why those measures matter for researchers and policy analysts.
Methods in brief: the authors used 10 search phrases across Google Web, YouTube, TikTok and four AI platforms (Gemini, ChatGPT GPT-4o, DeepSeek, Meta AI).
Quality and reliability were scored with a modified DISCERN (mDISCERN) and Global Quality Score (GQS).
Readability used the NHS Medical Document Readability Tool. Content was mapped across the cancer trajectory and inclusivity axes.
- Why this matters for researchers: heterogeneous source types (long-form websites, short-form video, LLM replies) require different evaluation signals (citation presence, balance, references, depth).
- Practical takeaway: automated triage plus structured thematic coding reveals where resources are trustworthy, shallow, or non-inclusive.
How to run an AI-enabled qualitative analysis on digital health resources
Overview
Overview: use an AI-enabled pipeline to ingest heterogeneous content, produce thematic and frequency analyses, compare segments, and output validated summaries.
Primary keyword: ai-enabled qualitative analysis.
Use an AI-enabled pipeline to ingest heterogeneous content (HTML, transcripts, LLM outputs, video captions), produce thematic and frequency analyses, compare segments, and output validated summaries.
Key steps (high level)
Key steps: harvest, normalize, auto-code and theme, check readability and inclusivity, compare segments, and deliver stakeholder-ready outputs.
1) Harvest: scrape search results and transcripts for websites, YouTube, TikTok, and capture LLM responses for fixed prompts.
2) Normalize: standardize text (remove boilerplate, keep citations), tag source metadata (author type, country, date).
3) Auto-code & theme: run AI-assisted code suggestions, then validate with a small manual sample.
4) Readability & inclusivity checks: measure estimated reading age, reading time, and presence of translation/cultural notes.
5) Cross-segment analysis: compare quality and themes by source type, country, and intended trajectory stage (diagnosis, treatment, end of life).
6) Deliver: produce stakeholder-ready visualizations (co-occurrence networks, hierarchical code trees, exemplar quotes).
Implications for research teams and policy analysts
For qualitative & UX researchers
For qualitative and UX researchers: include LLM outputs and short-form video transcripts in corpora and codebooks to reflect what users see.
Prioritise mixed-source corpora: include LLM outputs and short-form video transcripts in codebooks to reflect what users actually see.
Use combined metrics: pair mDISCERN-style reliability checks with thematic saturation and cross-segment frequency to spot gaps (for example, end-of-life content and inclusivity deficits).
For health policy & clinical leads
For health policy and clinical leads: measure both quality and readability before recommending resources, because high quality does not guarantee accessibility.
Don’t assume quality equals accessibility: high-quality websites scored better, but reading age (about 15 years) remains a barrier, measure both quality and readability before recommending resources.
Consider certification and regulated hubs to reduce harm from unvalidated AI outputs.
For data teams
For data teams: track provenance, save prompts and model versions, and automate guardrails to flag suspect outputs for review.
Track provenance and versioning: AI-generated outputs evolve rapidly, save prompts, model version, timestamps, and compare longitudinally.
Automate guardrails: flag outputs without citations or with likely hallucinations for manual review.
Do more, faster with Evidano
Problem: heterogeneous, noisy inputs
Evidano ingests web pages, transcripts, video captions and LLM replies, normalizes text and preserves metadata automatically.
Solution in Evidano: ingest web pages, transcripts, video captions and LLM replies, normalize text and preserve metadata automatically.
Problem: slow thematic synthesis
Evidano accelerates thematic coding and produces reproducible theme hierarchies and exemplar quotes for stakeholder reports.
Solution in Evidano: AI-assisted thematic coding, frequency analysis and co-occurrence networks that produce reproducible theme hierarchies and exemplar quotes for stakeholder reports.
Problem: readability and localization gaps
Evidano provides built-in readability metrics, translation with custom dictionaries, and rapid generation of alternative reading-level variants to test inclusivity.
Solution in Evidano: built-in readability metrics, translation with custom dictionaries, and rapid generation of alternative reading-level variants to test inclusivity.
Problem: iterative validation and stakeholder sign-off
Evidano supports collaborative workspaces, versioned codebooks, AI-chat over documents, and exportable visuals for policy briefs.
Solution in Evidano: collaborative workspaces, versioned codebooks, AI-chat over your documents to iterate hypotheses, and exportable visuals for policy briefs.
Security & compliance
Evidano stores data encrypted and does not use customer data to train third-party models, which matters for sensitive health materials.
Evidano uses encrypted storage and proprietary LLMs tuned for qualitative research. Your data is never used to train third-party models, critical when analysing sensitive health-related materials.
Checklist: reproduce this study in 10 steps (two-week pilot)
Follow this run-book to replicate Anand et al.'s comparison and extend it for your context in about two weeks.
- 1. Define the search phrases and platforms (use the paper’s 10 phrases as a starting point).
- 2. Harvest top N results per platform (for example, Google 50, YouTube/TikTok 20, AI single response per prompt).
- 3. Use Evidano to ingest raw HTML, video captions, and saved AI replies with metadata.
- 4. Auto-extract author type, country, date, and view/engagement metrics where available.
- 5. Run automated mDISCERN and GQS proxies (check for citations, balance, references).
- 6. Measure readability (reading age and time) across sources and flag materials above your target threshold.
- 7. Run thematic coding, then validate codes on a 10% random sample.
- 8. Produce cross-segment comparisons (source type × cancer-stage coverage × inclusivity).
- 9. Visualise results (word clouds, co-occurrence networks, hierarchical codes) and prepare a one-page executive brief.
- 10. Iterate with PPIE or stakeholder reviews and re-run targeted AI avatar interviews to fill content gaps.
FAQ: ai-enabled qualitative analysis
What did the Anand et al. PLOS Digital Health study find?
+
Direct answer: the study found Google web pages scored higher on quality metrics while AI outputs and TikTok scored lower, and written materials often required an estimated reading age of about 15 years.
The study screened 690 items and included 144 resources (Google web 39; AI 34; YouTube 24; TikTok 47), reported Google web mDISCERN 3.74 and GQS 3.72 versus AI mDISCERN 2.77 and GQS 2.32, and found mean reading times Google 11:45 min versus AI 02:09 min.
How can I reproduce the paper's comparison in my context?
+
Direct answer: reproduce the paper by harvesting comparable search results, ingesting content and AI replies, running automated quality and readability checks, and producing cross-segment analyses.
The post provides a 10-step checklist including harvesting top N results, ingesting HTML and transcripts, running automated mDISCERN and GQS proxies, measuring readability, and validating thematic codes on a 10% sample.
How does an AI-enabled workflow help detect inclusivity and readability gaps?
+
Direct answer: an AI-enabled workflow automates triage, measures readability (reading age and time), and enables cross-segment thematic comparisons that reveal inclusivity deficits.
Specifically, measure estimated reading age with standard tools, flag materials above your target threshold, and compare coverage across source types and cancer trajectory stages to spot gaps such as end-of-life coverage or lack of translation notes.
What practical role does Evidano play in this workflow?
+
Direct answer: Evidano ingests mixed-source content, preserves metadata, runs AI-assisted coding and readability checks, and produces reproducible visualisations and exportable reports.
Evidano automates ingestion of web pages, transcripts, and AI replies, supports collaborative versioned codebooks, and provides built-in readability metrics and visual outputs for stakeholder briefs.
Conclusion: what to do next
Conclusion: start with a small pilot that ingests a representative corpus, runs automated quality and readability checks, and produces cross-segment visualisations to reveal gaps and guide interventions.
Anand et al.’s July 9, 2026 analysis shows a clear quality hierarchy across platforms but highlights readable and inclusive gaps, precisely the problems AI-enabled qualitative analysis can expose and help fix.
If your team evaluates mixed-source health information or wants reproducible, auditable syntheses (including LLM outputs), start with a pilot: ingest a representative corpus, run automated quality and readability checks, and produce cross-segment visualisations.
Ready to try an end-to-end pipeline? See how Evidano operationalises these steps and accelerates reproducible qualitative research: Try Evidano for free.
Topics
- ai-enabled qualitative analysis
- qualitative analysis cancer resources
- digital health information
- readability inclusivity
Keep reading
- Commentary on NewsLARC/PM uptake in rural Bangladesh: Qualitative analysisActionable takeaways for researchers on barriers to LARC/PM in rural Bangladesh using qualitative analysis of contraceptive uptake, with AI research workflows and next steps.
- Commentary on NewsFaster Insights: AI-enabled Qualitative AnalysisHow AI-enabled qualitative analysis turns the PLOS One study on LARC/PM in rural Bangladesh into actionable themes and program recommendations. Learn the workflow.
- Commentary on NewsImplementing Nurse-Based Follow-Up: Qualitative InsightsActionable lessons from the PLOS One qualitative study on implementing nurse-based follow-up, with AI-enabled methods for faster thematic synthesis and decision-ready findings.