Hostname: page-component-6898cfcb55-cbk24 Total loading time: 0 Render date: 2026-11-11T10:16:12.145Z Has data issue: false hasContentIssue false
  • English
  • Français

Using generative artificial intelligence for outbreak communications: a case study of the 2026 Ebola outbreak

Published online by Cambridge University Press:  21 October 2026

Trini A. MathewOpen the ORCID record for Trini A. Mathew[Opens in a new window],
Kate M. Bowers and
David J. WeberOpen the ORCID record for David J. Weber[Opens in a new window]
Trini A. Mathew*
Affiliation:
Department of Internal Medicine, Oakland University William Beaumont School of Medicine , USA
Kate M. Bowers
Affiliation:
School of Engineering and Computer Science, Oakland University, USA
David J. Weber
Affiliation:
Division of Infectious Diseases of Infectious Diseases, The University of North Carolina at Chapel Hill, USA
*
Corresponding author: Trini A. Mathew; Email: tmathew@oakland.edu

Abstract

Communications during outbreaks and public health emergencies of infectious diseases can be challenging. Information provided by scientific organizations and federal agencies as well as international organizations may be misunderstood if not contextualized for specific audiences. Furthermore, lengthy statements and documents may be difficult to comprehend by the public, clinicians and others preparing and responding to outbreaks. We evaluated the use of generative artificial intelligence (AI) models to develop visual communications and infographics, using the recent 2026 Ebola outbreak as a case study. We identify the strengths and challenges of using AI models for producing outbreak communication materials.

Information

Type
Commentary
Information
Antimicrobial Stewardship & Healthcare Epidemiology , Volume 6 , Issue 1 , 2026, e344
Creative Commons
Creative Common License - CCCreative Common License - BYCreative Common License - NCCreative Common License - SA
This is an Open Access article, distributed under the terms of the Creative Commons Attribution-NonCommercial-ShareAlike licence (https://creativecommons.org/licenses/by-nc-sa/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the same Creative Commons licence is used to distribute the re-used or adapted article and the original article is properly cited. The written permission of Cambridge University Press or the rights holder(s) must be obtained prior to any commercial use.
Copyright
© The Author(s), 2026. Published by Cambridge University Press on behalf of The Society for Healthcare Epidemiology of America

Background

During a crisis and especially a public health emergency of international concern (PHEIC), 1,2 communications, clarity and connection with the public, scientists, physicians and clinicians is of utmost importance. The information shared needs to be succinct, accurate and tailored for the audience. 3 Frequently, information on the outbreak or topic du jour provided by the US Federal international agencies are densely populated with scientific words. Thus, one solution for sharing complex information is presenting visual abstracts, infographics with pictures.

Objective

We evaluated the information provided by generative artificial intelligence (AI) models to develop visuals and infographics on the current Ebola PHEIC declared on 17 May 2026, by the World Health Organization (WHO). 4 The objective was to determine the accuracy of the information provided in the end output of visual representation of the current known outbreak of the Ebola species Bundibugyo virus, to the prior outbreaks of the species of Zaire virus.

Methods

The free tiers of Microsoft Copilot 5 and Google Gemini 6 were used to generate comparative visual graphics on the two different known outbreaks by the Budibugyo and the Zaire species during two different sessions in late May 2026 to early June 2026.

Using Microsoft Copilot: On May 21st 2026, Microsoft Copilot was given a prompt to develop the infographic. The websites with scientific information provided by the US Centers for Diseases Control and Prevention (CDC) and WHO were put within the prompt (Figure 1). This prompt was general as noted in Figure 1 and stated “develop infographic comparing the Ebola Zaire with Bundibugyo, use the CDC websites,” with website links provided.

Figure 1.

Initial prompt on May 21st 2026.

Text box with instructions to develop an infographic comparing Ebola Zaire with Bundibugyo using CDC and WHO websites.

Microsoft Copilot verified the information received and furthermore, indicated its goal of only using accurate information (Figure 2).

Figure 2.

Response to the initial prompt, from Microsoft Copilot.

A list of sources and steps for creating an infographic on Ebola situation.

One of the coauthors had previously used this AI model to develop infographics related to infectious diseases, and with a previous interaction with this AI model, had requested assessment of the carbon footprint with generation of infographic. The estimation of the carbon footprint was not in the prompt; Microsoft Copilot opted to voluntarily report on this metric (Figure 3) stating “you previously requested carbon footprint transparency. I will include estimated CO 2 equivalent for generating the infographic; miles driven equivalent (per your prior format).” The importance of estimation or impact of the carbon footprint is not discussed as it was beyond the scope of this study.

Figure 3.

Response from AI model on prior request on carbon footprint generated by AI.

A note on carbon footprint transparency including estimated CO2 equivalent and EPA calculator link.

Further interactions of the AI model with the human (coauthor) is shown below (Figure 4). Although the human response was brief, this specific AI model provided a lengthy response. And of note, the retention of prior interactions of the human is seen by the response of the AI model, even indicating the known scientific conferences for infectious diseases.

Figure 4.

Response and suggestion from AI model to develop poster for known scientific conferences of infections diseases.

A text block describing a scientific poster layout guide for infectious disease and outbreak preparedness audiences.

The AI provided further choices and the prompt subsequently given by human (coauthor) at this juncture was “generate full visual poster mockup” as noted in Figure 5.

Figure 5.

Further responses from AI and prompt provided by human.

A list of options for generating a visual poster or presentation layout.

Use of AI models on a different date: In early June 2026, as the Ebola outbreak continued, further interactions with generative AI models were explored. A more specific prompt was used with Microsoft Copilot. However, although the AI provided texts and suggestions, it did not generate any visual infographics, stating the maximum limit was reached by the user (for the free tier).

Thus, another AI model (Google Gemini) was assessed to determine feasibility of developing community facing posters as well as infographics. The following specific prompt was provided along with the prior websites:

Generate a community-facing poster about the two species of ebola the zaire and bundibugyo based on these websites. Do not use any copyrighted phrases, nor icons, nor photos, nor pictograms https://www.cdc.gov/han/php/notices/han00530.html

https://www.who.int/news/item/17-05-2026-epidemic-of-ebola-disease-in-the-democratic-republic-of-the-congo-and-uganda-determined-a-public-health-emergency-of-international-concern

https://www.cdc.gov/ebola/hcp/clinical-guidance/index.html

https://wwwnc.cdc.gov/travel/notices/level3/ebola-democratic-republic-of-the-congo

A follow up prompt tailored for healthcare personnel was also provided with the aforementioned websites:

“Generate an infographic for healthcare personnel about the two species of ebola the zaire and bundibugyo based on these websites. Do not use any copyrighted phrases, nor icons, nor photos, nor pictograms”

and within a minute the infographic was generated by the Gemini AI model.

Of note, the difference in the prompts and AI models used, at different time periods, generated different visuals and infographics.

Results

Using Microsoft Copilot

Interaction in end of May: Over an initial period of 30 to 45 minutes the human (coauthor) interacted with the AI model providing the initial general prompt (“Develop infographic comparing the Ebola Zaire with Bundibugyo, using the websites links provided”). Revisions of this prompt were required with each output, with the human needing to respond to different options suggested as a solution by the AI model. Microsoft Copilot provided a visual abstract as shown in Figure 6. The AI visual output did provide an easy to understand visual comparison. It generated the similarities and differences on multiple points between these two species, in the domains of prior historical outbreaks, clinical, diagnostics, treatment and prevention.

Figure 6.

The first visual information shown with the errors including the CDC level 3 travel notice should have been under the column of Ebola Bundibugyo (generated May 21st 2026).

A table comparing Ebola Zaire and Ebola Bundibugyo virus species, including case fatality rates, major outbreaks, incubation periods, symptoms, transmission, diagnosis, treatment, vaccine, prevention, and travel notices.
Figure 6. Long description

The table compares Ebola Zaire and Ebola Bundibugyo virus species across various categories. It has two columns: Ebola Zaire and Ebola Bundibugyo, and ten rows: Virus Species, Case Fatality Rate, Major Outbreaks, Incubation Period, Symptoms, Transmission, Diagnosis, Treatment, Vaccine, Prevention, and Travel Notice. Row 1: Virus Species, Zaire ebolavirus, Bundibugyo ebolavirus. Row 2: Case Fatality Rate, 60-90 percent, 25-35 percent. Row 3: Major Outbreaks, DRC, West Africa 2014-16, Uganda 2007-2008, DRC 2012. Row 4: Incubation Period, 2-21 Days, 2-21 Days. Row 5: Symptoms, Fever, Vomiting, Hemorrhage, Fever, Vomiting, Rash. Row 6: Transmission, Blood, Body Fluids, Contaminated Objects, Handshake, Blood, Body Fluids, Contaminated Objects, Handshake. Row 7: Diagnosis, R T-PCR, Antigen Tests, R T-PCR, Antigen Tests. Row 8: Treatment, Inmazeb, Ebanga, Supportive Care. Row 9: Vaccine, r V S V-Z E B O V (Ervebo), No Approved Vaccine. Row 10: Prevention, Personal Protective Equipment, Hand Hygiene, Safe Burial Practices, Personal Protective Equipment, Hand Hygiene, Safe Burial Practices. Row 11: Travel Notice, C D C Level 3: D R C Outbreaks.

Subsequently, based on the multiple choices provided by the AI model to the coauthor led to another visual presentation (Figure 7). Of note, although initial assessment of the output can be impressive, upon closer review, it is noted that the output had spelling errors (“Epidomiologic”—instead of “Epidemiology”). Other misalignment of the visual representation includes the icon showing healthcare exposure did not truly represent the required personal protective equipment for healthcare personnel, nor provide explanation of the icons listed as “body fluids,” unsafe burials and healthcare exposure as the possible routes of exposure and thus risks of infection. Another major point of concern was that the visual infographic generated by the model highlights there - was genetic divergence between the two species of Zaire and Bundibugyo, however the percent divergence noted (55%–60%) was inaccurate.

Figure 7.

Visual representation generated by AI model (May 21st 2026).

Comparison of Ebola Virus Disease: Zaire vs. Bundibugyo.

As shown in Figure 8 the output generated by the AI model (21 May 2026) was in a poster format for a conference, with a hypothetical name (confabulated name, and university).

Figure 8.

Poster format output generated by AI model.

A comparison of Ebola Virus Disease between Zaire and Bundibugyo species, covering virology, epidemiology, clinical course, treatment, outbreak timeline, and geography.
Figure 8. Long description

Panel A: Virology & Epidemiology. The Zaire and Bundibugyo species show approximately 55-60 percent genome divergence. The case fatality rate for Zaire is 60-90 percent, while for Bundibugyo it is 25-35 percent. Transmission occurs through body fluids, funerals, healthcare settings, and community spread. Panel B: Clinical Course & Treatment. The clinical course is divided into dry and wet phases. The dry phase includes fever and fatigue, while the wet phase includes vomiting and diarrhea. Both phases can lead to leukopenia, thrombocytopenia, and elevated AST/ALT levels. Treatment options include Inmazeb & Banga and supportive care. Panel C: Outbreak Timeline & Geography. The timeline shows Zaire outbreaks from 1976 to 2022 and Bundibugyo outbreaks from 2007 to 2026. The map highlights outbreak locations in the Democratic Republic of the Congo (D R C) and Uganda.

Subsequently another session was initiated on the same day at the end of May. And a new prompt was provided to Microsoft Copilot including confirmation by the human to the AI for a community facing poster. As the prior websites were included, the AI output generated (Figure 9) information including the CDC logo. It also had the spelling error of “heading” instead of “headache.”

Figure 9.

Community facing poster generated by AI model in end of May 2026.

Poster about Ebola Virus Disease, its symptoms, spread, and prevention.

With the prompt “delete the words of CDC WHO Ministry of health and the CDC Logo” the AI output generated a revised poster (Figure 10), with also the correct spelling of headache.

Figure 10.

Community facing poster, without the logos, and spelling that AI self corrected.

A poster about Ebola Virus Disease (EVD), its symptoms, spread, and prevention.
Figure 10. Long description

Panel A: Description of Ebola. Ebola is a severe virus found in parts of Africa, with strains Zaire (Severe) and Bundibugyo (Milder). It does not spread through air. Symptoms include fever, weakness, vomiting, diarrhea, headache, muscle pain, and bleeding or bruising. Panel B: Symptoms and early care. Symptoms include fever, weakness, vomiting, diarrhea, headache, muscle pain, and bleeding or bruising. Seek medical help if feeling sick after travel. Panel C: How Ebola spreads. It spreads through body fluids, contaminated items, caring for sick, and funeral contact. It does not spread by air, food, or water. Panel D: Prevention and response. Wash hands often, avoid sick people, practice safe burials, undergo travel screening, and report symptoms.

The AI output was assessed on June 2nd, 2026 with a more specific prompt with the Microsoft Copilot. Although the AI model resulted in many rounds of text communications, the AI model declined to generate visuals indicating that the user (coauthor) had reached the limit for generating visuals.

Below are the results from the Gemini model from early June 2026 for the community facing poster (Figure 11). Although this output was generated within a minute and did not have spelling errors, the visual output provided lots of information that may have been overwhelming for some of the audiences. This AI model also developed visuals showing people with dark skin color, which may limit its use to certain audiences only.

Figure 11.

Different AI model output for community facing poster (on June 2nd 2026).

A poster about protecting the community from Ebola disease, detailing virus species, spread, symptoms, and community actions.

Results of the prompt for generating infographic specifically for healthcare personnel using the Gemini model are shown in Figure 12.

Figure 12.

Infographic for healthcare personnel generated by AI model (on June 2nd 2026).

Infographic comparing Ebola virus species for healthcare personnel, detailing clinical distinctions, infection prevention, and global outbreak context.
Figure 12. Long description

Infographic comparing Zaire Ebolavirus and Bundibugyo Ebolavirus. Panel A: Species Overview. Zaire Ebolavirus (EBOV) has caused most major outbreaks with significant virulence and pathogenesis. Bundibugyo Ebolavirus (BDBV) has less frequent outbreaks and is generally considered less severe. Core symptoms for both include fever, myalgia, vomiting, and diarrhea. Panel B: Clinical Distinctions. Observed differences include higher fatality rates, higher peak viral loads, and more reported neurological impairment for EBOV compared to BDBV. Panel C: Infection Prevention and Control. Strict adherence to PPE protocols is required for all Ebola virus disease suspect cases. Diagnostic considerations include high sensitivity and specificity with common assays for EBOV and may require specific PCR primers or assays for BDBV. Panel D: Global Outbreak Context. Major Zaire outbreaks occurred in the Democratic Republic of the Congo and Uganda, with recent Bundibugyo outbreaks identified in the same regions.

Limitations

Generative AI has a variety of limitations. The figures shown of the initial prompts and outputs generated primarily demonstrate the limitations associated with prompt writing, and the results generated with vague as well as more specific prompts. Other limitations include model/tool choice and tier choice. For example, Microsoft Copilot is not a highly rated tool for image generation compared to other tools such as Google Gemini. Furthermore, even within a single tool such as Google Gemini, the model choice (eg, fast vs thinking) will significantly impact the accuracy of the image generated. Finally, many features are locked behind different payment tiers; the figures above demonstrate Microsoft Copilot’s free tier. Higher payments may yield different models due to access to more recent (and likely more intelligent) model options. Users also need to factor the rapidly evolving nature of AI models and the limited generalizability of infographics, especially with any new scientific developments leading to changes in generated output.

Discussions

Although the clinician needs to be aware of the aforementioned limitations, there are also advantages of using generative AI to assist with developing communication infographics. Using generative AI can assist with communicating complex information in simple visual formats. Time spent on designing visuals can also be decreased with the use of generative AI. Especially as outbreaks unfold, and there is change in outbreak numbers and information received, and scientific studies published, developing more timely visuals, in simple formats, can assist with public health messaging.

Conclusion

Clinicians can leverage existing generative AI models to assist with developing infographics and other visuals for educational purposes. These models can be used to assist contextualize the information for the intended audience. However, clinicians need to also know the limitations of these models, as well as providing specific prompts. As these models get further trained by the interactions of multiple users, the output generated may be unique. Ultimately, although clinicians may adopt and use generative AI, it is wise to verify the visual output for both scientific accuracy as well as content representation. This work does not include formal assessment of usability, readability, user satisfaction or communication effectiveness. Further assessments are needed to understand implications of widescale adoption of generative AI for communications on infections and public health risks.

Author contributions

Manuscript conceptualization: TAM, DJW

Methods and results: TAM, KB, DJW

Manuscript writing and editing: TAM, KB, DJW

Financial support

No funding received for work and developing this manuscript.

Competing interests

KB and DJW do not have any conflicts of interest to report. TAM reported individual stocks in Alphabet Inc. and GE Healthcare Tech Inc.

References

World Health Organization. Emergiencies: International health regulations and emergency committees. https://www.who.int/news-room/questions-and-answers/item/emergencies-international-health-regulations-and-emergency-committees.Accessed, June 2nd 2026.Google Scholar
World Health Organization. 2005. International Health Regulations, 2005. Published 2005, amended in 2014 and 2022. https://apps.who.int/gb/bd/pdf_files/IHR_2022-en.pdf. Accessed, June 8th 2026.Google Scholar
Crisis and Emergency Communications (CERC). Messages and Audiences. U.S. Department of Health and Human Services, Centers for Disease Control and Prevention. https://www.cdc.gov/cerc/media/pdfs/CERC_Messages_and_Audiences.pdf. Accessed, June 2nd 2026.Google Scholar
World Health Organization Epidemic of Ebola Disease caused by Bundibugyo virus in the Democratic Republic of the Congo and Uganda determined a public health emergency of international concern. https://www.who.int/news/item/17-05-2026-epidemic-of-ebola-disease-in-the-democratic-republic-of-the-congo-and-uganda-determined-a-public-health-emergency-of-international-concern. Accessed June 2nd 2026.Google Scholar
Microsoft Copilot. https://copilot.microsoft.com/. Accessed May 21st and June 2nd 2026.Google Scholar
Google Gemini. https://gemini.google.com/app. Accessed June 2nd 2026.Google Scholar
Figure 0

Figure 1.Initial prompt on May 21st 2026.

Figure 1

Figure 2.Response to the initial prompt, from Microsoft Copilot.

Figure 2

Figure 3.Response from AI model on prior request on carbon footprint generated by AI.

Figure 3

Figure 4.Response and suggestion from AI model to develop poster for known scientific conferences of infections diseases.

Figure 4

Figure 5.Further responses from AI and prompt provided by human.

Figure 5

Figure 6.Figure 6 long description.The first visual information shown with the errors including the CDC level 3 travel notice should have been under the column of Ebola Bundibugyo (generated May 21st 2026).

Figure 6

Figure 7.Visual representation generated by AI model (May 21st 2026).

Figure 7

Figure 8.Figure 8 long description.Poster format output generated by AI model.

Figure 8

Figure 9.Community facing poster generated by AI model in end of May 2026.

Figure 9

Figure 10.Figure 10 long description.Community facing poster, without the logos, and spelling that AI self corrected.

Figure 10

Figure 11.Different AI model output for community facing poster (on June 2nd 2026).

Figure 11

Figure 12.Figure 12 long description.Infographic for healthcare personnel generated by AI model (on June 2nd 2026).