By Ayan Chakraborty, Founder, SBA Innovation Pvt Ltd, Kolkata. Last updated . Reviewed by Ayan Chakraborty.
How can an NBFC get cited by ChatGPT?
When ChatGPT searches the web it retrieves pages through its search crawler and quotes passages that answer the question cleanly. So the work is practical: make sure OAI-SearchBot can fetch your pages, that the key text is in the HTML rather than behind scripts, and that each important question has one direct answer with the number, rule or definition in plain text. Without search switched on, ChatGPT answers from training data, so a brand mention there depends on model updates and cannot be scheduled.
Treat it as a measurement problem as well. Build a fixed set of borrower prompts, run each several times on each assistant, log whether you are cited with a link, mentioned without one or absent, and repeat on a schedule. Nobody can guarantee a citation, because answers vary by run and by model update; the plan raises your odds and shows what moved.
- Allow OAI-SearchBot and PerplexityBot, and check that your CDN or WAF does not block them.
- Put fees, rates, eligibility and documents in plain HTML, not images or JavaScript-only tables.
- Write one question per section with a direct answer in the first 40 to 60 words.
- Add Organization and Article schema and a visible author and updated date.
- Use the same name, 'SBA Innovation Pvt Ltd, Kolkata' style, in every bio and listing.
- Track a fixed prompt set three times per assistant and repeat on days 0, 28, 56 and 90.
How do I get named when someone asks ChatGPT which personal loan lender to trust?
Recommendation prompts lean on what the wider web says about you, not only on your own site. Assistants draw on reviews, comparison pages, news, directories and aggregators, so being named is mostly a result of being consistently and accurately described in several independent places, alongside having clear facts on your own pages.
Make your own facts unambiguous first: legal name, RBI registration, products, rate range and fees in readable HTML. Then earn independent mentions honestly: verified customer reviews with consent, accurate directory profiles and coverage of original data. Fake reviews and mention farming do not work and can be discounted. Expect this to take months and to stay uncertain: in our own work some comparison prompts still went to aggregators, and we do not expect to win them.
| Source type | What it gives the assistant | What you can do |
|---|---|---|
| Your own pages | Facts, fees, eligibility, registration | Plain HTML, direct answers, schema |
| Review platforms | Customer sentiment and volume | Ask real customers, with consent |
| Directories and listicles | Category and comparison context | Complete accurate profiles |
| News and research | Third-party credibility | Publish original data and pitch it |
| Aggregators | Rates and fees, sometimes outdated | Correct stale facts at source |
From our client work: 300 prompts, five assistants
For a consumer lending NBFC we ran a 300-prompt borrower set across ChatGPT, Google AI Overviews, Gemini, Perplexity and Copilot. Over eight months, prompts citing the lender with a link rose from 11 to 64, prompts mentioning it from 42 to 102, and AI share of voice against six competitors from 8% to 27%.
Method: Engagement 031: median of three runs per assistant in the same week, fixed competitor list and scoring rubric, run from a neutral India-located session with no chat history.
What did not work: Long 2,500-word guides did almost nothing; assistants kept quoting a competitor's 90-word fee summary until we rewrote for extractability.
Read case study 031 Prompt set, baseline, assumptions and footnotes.