Most people try Andi Search the way they try a new restaurant: they order the thing they always order, decide it’s fine, and never come back. They type “best laptop 2025”, skim the reply, shrug, and return to Google. That’s a waste of a genuinely different tool.
Andi doesn’t behave like a keyword box, and once you change two or three habits the answers get noticeably sharper. Here’s the workflow I’ve settled into after a few months of using it as my first stop for anything I’d otherwise burn twenty minutes tab-hopping to answer.
Be clear about what Andi actually is
Andi is an independent search engine built around a chat window rather than a page of ten blue links. No ads wedged above the results, no behavioural profile assembled from your queries, and every answer arrives with sources attached so you can see where a claim came from. The team behind it has been fairly explicit that the whole point is answering before you click anything, which describes both its strength and its limit.
In practice, that means your job stops being find pages about this and becomes describe what I actually want to know. That one shift drives everything below.
Step 1: Write a sentence, not a search string
Keyword habits die hard. Ask Andi “best coffee grinder 2025” and you’ll get a competent but predictable roundup of the usual suspects. Rewrite it as a situation and the answer changes shape entirely:
“I make two flat whites a day, I’m upgrading from a cheap blade grinder, and I’d rather not spend more than £120. What should I look at, and what’s the one difference I’ll actually notice?”
That version forces a recommendation against real constraints. Instead of five grinders, you get a shortlist of two, a plain-English explanation of burr versus blade, and a note that a hand grinder may beat anything electric at that price. Same tool, same underlying index, completely different answer.
The three ingredients worth including
- Context — who you are and what you’re doing with the answer.
- Constraints — budget, deadline, location, gear you already own.
- What “good” means to you — cheapest, quietest, fastest, least maintenance.
You don’t need all three every time. But when a response feels generic, it’s almost always because the question was generic.
Step 2: Chain follow-ups instead of starting over
Andi holds the thread, so your second question can lean on the first. This is where it separates itself from a traditional search session, where narrowing a question means retyping it with extra words and hoping for the best.
Real follow-ups from that grinder conversation:
- “Which of those two is quieter in a small flat?”
- “What do owners complain about after six months?”
- “If I stretched to £180, what actually changes?”
Each one narrows without losing the original constraints. The replies get shorter and more specific as you go, which is exactly what you want. Four or five turns in, you usually have a decision rather than a reading list.
Step 3: Read the sources on anything you’ll act on
Andi cites where its answers come from, and it’s worth clicking through on claims that carry weight — a price, a warranty term, a safety or health statement. Generative search tools can compress or misread a source, and no engine is immune to that. Treat the citation as a starting point, not a receipt.
That habit matters more every year. The big engines keep pushing ordinary search results further down under AI-generated summaries, which means more of what you read online is a synthesis of something rather than the something itself. Checking one link occasionally keeps you calibrated.
Step 4: Ask for trade-offs, not just recommendations
The best questions in a conversational engine are the ones that expose disagreement. Three phrasings that reliably beat “what’s the best X”:
- “What would make you recommend the opposite?”
- “Where do reviewers disagree on this?”
- “What’s the strongest argument against the option you just suggested?”
A recent search about switching broadband providers is a good example. Asking where reviewers disagreed surfaced that most speed complaints traced back to one specific installation type — detail the first three answers had quietly smoothed over.
Step 5: Know where Andi isn’t the right tool
Debugging and code
Stack traces, library quirks, version conflicts: a developer-focused engine is faster and less likely to hand you confident nonsense. Phind was built for exactly that frustration, and it shows in how it formats snippets.
Live breaking news
Anything that happened in the last hour is a coin flip on any AI-first engine, including this one. Go to a news site for that.
Long, multi-source research
If a question needs twenty sources compared and footnoted, a research-oriented assistant will serve you better — Perplexity has quietly become the default for that kind of work for good reason. None of this is a knock on Andi. Picking the right tool beats optimising prompts inside the wrong one.
A full session, start to finish
Here’s what a real run looks like, using a question I actually had: three days in Lisbon in October with a two-year-old.
Turn 1: “We’ve got three days in Lisbon in October with a toddler, staying near Príncipe Real. We don’t want to be constantly on the move. What’s a realistic daily plan?” Back comes a neighbourhood-level itinerary, with a warning about hills and cobbles.
Turn 2: “Which of those bits is hardest with a pushchair?” The answer reorders the days and drops one stop entirely.
Turn 3: “Best time of day to do Belém without the crush?” Specific, time-bound, useful.
Turn 4: “What’s the one thing locals would tell me to skip?” This is the turn that produced something no guidebook had mentioned.
Turn 5: “Turn that into a three-day list with rough walking times.” Roughly four minutes of typing, and I had something I could screenshot and use.
Five prompt patterns worth stealing
- Constraint stacking: “I have X, I need Y, I can’t do Z. What are my options?”
- Ask for the trade-off: “What am I giving up if I choose the cheaper one?”
- Ask what would change the answer: “What piece of information would flip your recommendation?”
- Ask for the other side: “Who disagrees with this, and why?”
- Ask for compression: “Summarise that as five steps I can follow tonight.”
Those five patterns cover most of what you’ll ever need, and they work on any conversational search engine, not just this one. The prompt is the skill. Everything else — the interface, the model, the index — is just plumbing you happen to be renting for a few minutes.