How to brief an AI shopping assistant for better results

An AI shopping assistant can save you a great deal of time, but like any good helper it works best with a clear brief. The difference between "I need a new kettle" and "I need a quiet 1.7 litre kettle under £50 that matches a white kitchen, delivered by Friday" is the difference between a long list of guesses and three options worth considering.

This guide covers what to include, how to refine the results and how to make the final decision with confidence. The examples use Ottr, but the principles apply to any assistant.

Start with the job, not the product

It is tempting to name a product straight away. Often it is more useful to describe the job you need done.

• Instead of "a 65 watt charger", try "a charger for my 14-inch laptop that I can also use for my phone".
• Instead of "a stand mixer", try "something to make bread dough twice a week without wearing out".

Describing the job lets the assistant suggest things you might not have thought of, and spot when the product you had in mind is not the best fit.

The five details that matter most

You do not need to write an essay. In most cases, five details do the heavy lifting.

1. Budget. A ceiling ("under £40") or a range ("£80 to £120"). Say whether delivery is included.
2. Timing. When you need it by, if it matters. "By Saturday" rules out a lot of options quickly.
3. Hard requirements. Size, compatibility, dimensions, materials, anything that is non-negotiable.
4. Preferences. Things you would like but could live without: colour, brand, finish.
5. What to avoid. Brands you have had bad experiences with, materials you dislike, features you do not need.

A brief covering those five might read: "Walking boots, women's size 6, waterproof, under £120 including delivery, needed before 18 October. Prefer a wider fit. Not suede."

Be specific about compatibility

Compatibility is where shopping goes wrong most often: the cartridge that does not fit, the cable with the wrong connector, the cover for last year's model.

If something needs to work with something else, give the assistant the precise details:

• the make and model number of the device
• the part number printed on the item you are replacing
• measurements, if fit matters

If you are not sure, say so. A good assistant should ask a clarifying question rather than guess, and Ottr is designed to do exactly that.

Say how you will choose

Tell the assistant what matters most when the options compete. Are you after the lowest price, the fastest delivery, the most durable option, or the best reviewed? "I would rather pay a bit more for something that lasts" changes the shortlist considerably.

It also helps to say how many options you want to see. Three is usually plenty. More than five tends to recreate the browsing you were hoping to avoid.

Refine in conversation

The first shortlist is a starting point, not a final answer. Because you are messaging, refining is quick:

• "The second one, but in black?"
• "Anything similar that arrives sooner?"
• "Why is the first one so much cheaper?"
• "Drop anything with a plastic jug."

Ottr remembers the conversation, so you can refer back to earlier options without repeating the details. That works across channels too: you can start on the web and carry on in Telegram.

Ask the questions you would ask a shop assistant

You are entitled to understand a recommendation. Useful questions include:

• What are the trade-offs between these options?
• Who is the retailer, and what is their returns policy?
• Is delivery included in that price?
• Is this in stock now, or on back order?

If an assistant cannot answer something, it should say so plainly rather than fill the gap with guesses.

Check the approval page properly

When you are ready, Ottr prepares the purchase and sends you a link to a secure approval page. This is the one step where you should slow down for a moment. Check:

• the exact item, including size, colour and quantity
• the total, including delivery
• the retailer
• the delivery address
• the card that will be charged

The approval is tied to exactly those details. If the price changes before you approve, the approval will not go through, and you will be told that nothing has been charged. You can read why we built it this way in Why a 'yes' in a chat should never be enough to buy something.

A few things to avoid

• Do not share card numbers in chat. Ottr will never ask for them. Cards are added securely through our payment provider, and each purchase is paid with a single-use virtual card.
• Do not approve in a hurry. The link lasts 15 minutes, and you can always ask for a new one.
• Do not assume the assistant knows everything. It can only compare what it can reach. If you have a particular retailer in mind, mention it.

Example briefs

Here are a few briefs that tend to get good results.

I need a present for a nine-year-old who loves space. Budget about £30. Not a book, she has plenty. Needs to arrive by the 24th.

Replacement water filter for a fridge. The old one says DA29-00020B on the side. Two-pack if it is cheaper per filter.

A rucksack for a daily cycle commute. Fits a 15-inch laptop, waterproof or comes with a cover, under £90. I prefer something plain, no big logos.

Each one states the job, the budget, the timing or constraint, and at least one preference.

A note on where Ottr is today

OTTR is in development. You can brief Ottr on the web now, and on Telegram where it is enabled, with WhatsApp waiting on Meta's approval. Purchases currently run in a sandbox against a test merchant, so the options you see and the orders you place are for testing only. It is a good way to get a feel for how briefing and approving work before purchasing goes live.

The takeaway

Good results come from good briefs. Describe the job, give your budget and timing, name the must-haves and the deal-breakers, and refine in conversation. Then take a moment on the approval page to check that what you are approving is exactly what you want.

Ready to try it? Get started with Ottr. If you are curious how assistants differ from the chatbots you may be used to, read What is an AI shopping agent?, and for why messaging suits shopping so well, see Why the next shopping app is a conversation.