IgeraFincas vs ChatGPT: The Difference Residents Notice

The experiment: one question, two systems, a manager as judge

We asked three property managers with over ten years of experience to evaluate the responses of two systems to the same question. They were not told which system was which. They were only asked to rate legal accuracy, source citation and practical utility on a scale of 1 to 10.

The question: "An owner with reduced mobility requests the installation of an elevator in a four-story building without an elevator. The community has 8 owners. How many votes does it need and who pays?"

Response from System A (ChatGPT-4o)

"For the installation of an elevator in a community of owners in Spain, according to article 17.2 of the LPH, the affirmative vote of 3/5 of the owners and participation shares is required. In your case, with 8 owners, you would need at least 5 favorable votes. As for payment, all owners contribute according to their participation share, although the flats on the top floor usually have higher shares for these improvements."

Manager evaluation: legal accuracy 3/10, source citation 2/10, practical utility 4/10. The quorum cited (3/5) is incorrect for this scenario. The statement about top floor apartments is invented.

Response from System B (IgeraFincas RAG)

"According to Art. 10.1.b of the LPH (current text loaded: 01/15/2026): It is the responsibility of the community, even without prior meeting agreement, to carry out accessibility works for people with disabilities or over 70 years old, when the annual amount to be charged for these, after deducting subsidies, does not exceed twelve monthly ordinary common expense payments. If it exceeds that amount, it requires meeting agreement by simple majority (Art. 17.2). As for payment, Art. 9.1.e establishes that all owners contribute according to their participation share. Note: if the requesting owner has recognized disability of at least 33%, the community cannot refuse to carry out the works, regardless of cost. [Source: LPH_consolidated_2024.pdf, p. 8 and 18]"

Manager evaluation: legal accuracy 9/10, source citation 10/10, practical utility 9/10. Exact citation, distinction between mandatory and discretionary works, and a practical nuance that system A completely omitted.

Four structural differences

1. Source of information

ChatGPT generates statistically probable text. IgeraFincas retrieves text from actual loaded documents with date and version.

2. Traceability

ChatGPT's response cannot be quickly verified. IgeraFincas' response includes document name and page number. Verification in 15 seconds.

3. Community specificity

ChatGPT doesn't know the bylaws of the Calle Balmes 47 community. IgeraFincas does, because they're loaded. The response can include specific clauses from that community.

4. Regulatory updates

ChatGPT has a training cutoff date and doesn't know if the LPH has been modified after that. IgeraFincas updates every time a new regulatory document is loaded, with visible date.

When to use ChatGPT and when to use IgeraFincas

ChatGPT is useful for writing tasks, summarizing your own texts, drafts of informal communications or brainstorming. It is not appropriate for legal queries where source precision is critical.

IgeraFincas is appropriate for answering owner questions about regulations, bylaws, minutes, contracts and any document that has legal or financial consequences.

Frequently asked questions

Can I use both systems in parallel?

Yes, and many managers do. ChatGPT for communication drafts, IgeraFincas for owner queries that require legal accuracy.

Does IgeraFincas use ChatGPT behind the scenes?

No. IgeraFincas uses Gemini 2.0 Flash as the generative model, with the RAG pipeline on Supabase pgvector. The model can only generate text based on retrieved document fragments, not its general training.

Are there cases where IgeraFincas also cannot respond?

Yes. When the response requires complex legal interpretation, assessment of specific case circumstances, or when the relevant document has not been loaded. In those cases, the system informs the owner that the query will be handled by the manager, and registers the question for follow-up.

About the author

Dr. Albert Rovira

Head of AI Research, Igera Solutions

Researcher in Large Language Models with a PhD in Natural Language Processing. Has published research on hallucinations in LLMs and safe architectures for critical applications. Scientific foundation behind IgeraFincas.

Related Articles

• The Future of Property Management in Spain: AI Trends 2026-2030
• WhatsApp for Communities: How to Manage 850 Chats Without Chaos
• How AI Prepares the Manager for the Owners Meeting
• 24/7 Chatbot for Communities: Resolving Emergencies at 3 in the Morning
• AI vs. Traditional Manager: Competition or Alliance?