When I run Deep Research query, I get visually appealing output, which I can (with use of external tools) convert to e.g. PDF. Unfortunately, if I request output in some structured form (e.g. latex citations with bibtex references), the algorithm obeys, provides me with latex file and bibtex file (yay!), but the references are messed up, they are in some kind of OpenAI internal form that does not get properly resolved!
For example (see italic in the below latex code)
Ovens are used daily in EU households for everything from quick meals to traditional Sunday dinners. A recent pan-European analysis found that Northern, Southern, and Western Europe lead the world in home-cooked meal frequency (about 7.8 home-cooked meals per week on average)**​:contentReference[oaicite:0]{index=0}**, and many of those meals involve the oven. Busy lifestyles have also driven the popularity of convenient oven-ready foods (like frozen items), while long-standing culinary traditions keep classic
Chatbot is unaware of this, actually, he is pretty unaware of the previous deep search, and even if I copypaste the output (bibtex and latex) to the next prompt and use o3 for example to fix it, it will start hallucinate heavily (bunch of links with example dot com).
I am putting this down to a bug, maybe it is “missing functionality” but I think it is at least oversight in implementation
Below is the bibtex file coming from deep research:
@article{IndexBox2020,
author = {IndexBox},
title = {Britons Consume the Most Frozen Potatoes in the EU, nearly 70\% Comes from the Netherlands and Belgium},
journal = {Global Trade Magazine},
year = {2020},
url = linkremoved due to posting rules
}
@article{GrandView2020,
author = {{Grand View Research}},
title = {Europe Frozen Pizza Market Size \& Outlook, 2019-2027},
year = {2020},
url = linkremoved due to posting rules,
note = {Industry market report}
}
@misc{WPR2025,
author = {{World Population Review}},
title = {Pizza Consumption by Country 2025},
year = {2023},
url =linkremoved due to posting rules
}
@article{BlackpoolGazette2021,
author = {{Blackpool Gazette}},
title = {Survey reveals what we love most about our roast dinners},
year = {2021},
url = linkremoved due to posting rules
}
@article{Fernandes2025,
author = {Janice Fernandes},
title = {How people in the UK prefer to cook: From scratch or meal kits?},
year = {2025},
journal = {YouGov (business.yougov.com)},
url = linkremoved due to posting rules
}
@article{Wall2018,
author = {Denise Wall},
title = {Baked salmon, mince-macaroni casserole top list of favourite foods in Finland},
year = {2018},
journal = {Yle News},
url = linkremoved due to posting rules
}
@article{IPO2021,
author = {{International Pasta Organization}},
title = {Lasagna: The most-loved pasta dish that brings people together},
year = {2021},
journal = {International Pasta Organisation News},
url =linkremoved due to posting rules
}
@article{Garget2020,
author = {Jacqueline Garget},
title = {The world\textquotesingle s their fish finger},
year = {2020},
journal = {EIT Food News (reprinted from University of Cambridge research)},
url = linkremoved due to posting rules
@article{AHDB2023ready,
author = {{Agriculture and Horticulture Development Board (AHDB)}},
title = {Are cost conscious consumers still reaching for ready meals?},
year = {2023},
journal = {AHDB Consumer Insight News},
url = linkremoved due to posting rules
}
@article{AHDB2024baking,
author = {{Agriculture and Horticulture Development Board (AHDB)}},
title = {Baking in 2023: Cakes are the rising star},
year = {2024},
journal = {AHDB Consumer Insight News},
url =linkremoved due to posting rules
}
@misc{LeGratin2014,
author = {{L\textquoteright{}\'equipe du gratindauphinois.fr}},
title = {Le gratin dauphinois parmi les plats pr\'ef\'er\'es des Fran\c{c}ais},
year = {2014},
url = linkremoved due to posting rules
note = {Reporting TNS Sofres survey (2011)}
}