Evidano is an AI-powered qualitative data analysis platform that handles OCR, multilingual inputs, and AI-assisted coding to scale thematic analysis. Researchers and policy teams tracking corporate climate messaging face two problems: large, multilingual corpora (annual reports, filings, PR) and subtle rhetorical shifts, greenwash that looks like progress. This week’s research roundup (posted 16 July 2026) highlights a textual analysis of four major fossil fuel companies showing strategic shifts in renewable-energy messaging between 2016–2022. In this post you’ll learn a concise, reproducible workflow for running qualitative analysis of fossil fuel discourse and how Evidano speeds the path from raw reports to stakeholder-ready themes, frequency counts, and cross-segment comparisons.
Key Takeaways
Desai et al.'s document-level comparison of annual reports (2016–2022) shows renewable-energy narratives often mask operational positions that delay climate action, summarized in the Skeptical Science roundup on 16 July 2026.
This post provides a reproducible, AI-assisted workflow to extract themes, quantify messaging shifts across 2016–2022, and export stakeholder-ready evidence for research, policy, and communications teams.
- Desai et al. compared annual reports from ExxonMobil, BP, Shell, and TotalEnergies for 2016–2022 and documented increased renewable-energy messaging alongside lagging operational stances.
- Scalable qualitative methods combine OCR, codebook-driven coding, frequency and co-occurrence analysis, and temporal comparison to surface divergence between rhetoric and action.
- Evidano enables OCR, multilingual inputs, AI-assisted coding, co-occurrence networks, and time-series exports to reproduce the analysis and produce stakeholder-ready outputs.
Findings snapshot
| Item | Value | Source / Note |
|---|---|---|
| Corpus | Annual reports from 4 firms (ExxonMobil, BP, Shell, TotalEnergies), 2016–2022 | Desai et al.; summarized in Skeptical Science, 16 July 2026 |
| Primary finding | Nuanced renewable-energy messaging increased while operational stances lagged | Desai et al., Energy Sustainability and Society (DOI link in Skeptical Science summary) |
| Public opinion context | 68% of Americans say global warming is happening (Yale/GMU, Spring 2026) | Yale/George Mason report cited in Skeptical Science roundup |
| Implication for analysts | Need for theme frequency tracking, temporal comparison, and cross-company code alignment | This post maps an AI-enabled qualitative workflow |
What happened & how the analysis works
Desai et al. used document-level textual analysis of annual reports to compare external messaging about renewables against inferred strategic positions. Methods like this generally combine close reading with coding (manual or assisted), timeline comparison (2016–2022), and keyword/theme frequency to surface divergence between rhetoric and action.
- Data inputs: corporate annual reports (PDFs), public statements, and timelines (2016–2022).
- Analytic steps: corpus ingestion → preprocessing (OCR, cleanup) → codebook creation → thematic coding → frequency & co-occurrence analysis → temporal comparison.
- Validation: triangulate textual signals with investment, asset portfolios, and policy activity where available.
So what for researchers and analysts
For investigative researchers
Investigative researchers can use reproducible coding to show where messaging diverges from investment choices. Use reproducible coding to quantify how often "renewable" appears alongside caveats or hedges, and produce exportable quote packs for FOI or press teams.
For policy teams & regulators
Policy teams and regulators can produce time-series evidence of shifting frames to support rulemaking. Produce time-series evidence of shifting frames and use cross-segment analysis (by company, year, region) to target accountability measures.
For communications and risk teams
Communications and risk teams can detect greenwash patterns early by monitoring co-occurrence networks. Detect greenwash patterns (for example, "net-zero" plus financing language) and translate findings into stakeholder-ready dashboards and short memos.
Do more, faster with Evidano
Ingest messy corpora
Evidano ingests messy corpora and handles OCR, language detection, and multilingual inputs so you can work with the full 2016–2022 corpus without manual rekeying. Upload annual reports (PDF, DOCX), scraped press releases, or CSV survey responses and let Evidano standardize them.
Scale thematic and frequency analysis
Evidano runs automated theme extraction and frequency counts, then lets teams refine results with human-reviewed codebooks. Run automated theme extraction and frequency counts, then refine with a human-reviewed codebook to produce hierarchical code → subcode visualizations and co-occurrence networks.
Compare segments and timelines
Evidano supports cross-segment analyses to show where rhetoric deviates from peers over time. Run company vs company and year-on-year comparisons and export time-series tables for policy appendices or briefings.
Secure, reproducible, research-focused
Evidano keeps data encrypted and does not use customer data to train third-party models. Import codebooks, run AI-assisted coding, and save reproducible analysis pipelines that auditors or reviewers can re-run.
Checklist: 7-step workflow to reproduce the Desai-style analysis
Follow these seven steps to reproduce the Desai-style analysis using annual reports from 2016–2022.
Step 1: Collect PDFs of annual reports (2016–2022) and standardize file names.
Step 2: Run OCR and language cleanup; remove boilerplate sections you don't want coded.
Step 3: Seed a codebook from a manual read (e.g., "renewable framing", "investment hedging", "policy deflection").
Step 4: Upload corpus to Evidano and run AI-assisted coding across the codebook; review and correct a 10% sample.
Step 5: Generate frequency tables, co-occurrence networks, and timeline charts to measure framing shifts.
Step 6: Cross-check textual signals with investment data or public filings (spot discrepancies).
Step 7: Package quotes, visuals, and an executive memo for stakeholders.
FAQ: qualitative analysis of fossil fuel discourse
How do I ensure coding is reliable across companies?
Ensure coding is reliable by using a shared codebook, running intercoder checks, and auditing samples. Use a shared codebook, run intercoder checks on a sample, and leverage Evidano's AI-assisted coding to apply consistent labels at scale; then audit a stratified sample for drift.
Can I compare rhetoric to actual investment behavior?
Yes, you can compare rhetoric to investment behavior by merging theme frequency timelines with investment or asset data. Export theme frequency timelines and merge them with investment or asset data (CSV); Evidano supports cross-segment joins so you can test correspondence between language and actions.
Is multilingual analysis supported?
Yes, multilingual analysis is supported through translation and custom dictionaries. Evidano supports translation with custom dictionaries, which is useful if reports or press releases include non-English sections or localized phrasing.
Wrapping up & next steps
Desai et al.'s 2016–2022 comparison shows why qualitative, reproducible analysis matters: rhetoric can be weaponized to delay action. If your team needs to scale this kind of analysis (across companies, years, and languages) start with a two-week pilot that follows the 7-step checklist above.
- Ready to try it? See how Evidano ingests reports, runs thematic + frequency analyses, and exports stakeholder-ready visuals: Try Evidano for free.
- Primary source: full roundup and links at Skeptical Science (16 July 2026).