How is an AI assistant different from search? Search points you to sources. An assistant writes an answer, which you must then check.
Short answer. Search returns a ranked list of documents for you to read; an AI assistant writes a fluent answer in its own words. Google describes search as crawling pages, indexing them, and serving results that are relevant to a query. An assistant answering from memory can be wrong without showing it: an audit of four generative search engines found that only 51.5% of generated sentences were fully supported by their citations. A good office setup combines the two: an assistant that first retrieves passages from your own documents, as in retrieval-augmented generation, then answers from them and shows the passage so you can check it.
01
Side by side
Question
Search
Assistant answering from memory
Assistant with retrieval
What you get
A ranked list of pages or files
A written answer
A written answer plus the passages it used
Where the facts come from
The documents themselves
The model’s training, which may be stale or wrong
Documents retrieved at question time
How you check it
Read the result
Search for the claim yourself
Open the passage shown beside the answer
Main risk
Missing the right document, or reading too much
Confident, invented detail
An irrelevant passage retrieved, or an answer that goes beyond it
Suits
Finding an exact clause, file or date
Drafting and explaining general topics
Summarising and answering from your own records
02
What the evidence says about assistants
In Liu, Zhang and Liang’s 2023 audit of Bing Chat, NeevaAI, Perplexity and YouChat, only 51.5% of generated sentences were fully supported by citations, and only 74.5% of citations supported their sentence.
In the European Broadcasting Union and BBC study of more than 3,000 answers on news from ChatGPT, Copilot, Gemini and Perplexity, almost half had at least one significant issue and a third had serious sourcing problems.
Retrieval helps but is not a cure: Stanford found legal research tools built on retrieval hallucinated between 17% and 33% of the time.
The 2020 paper that introduced retrieval-augmented generation combined a language model with a searchable index to produce more specific and factual output than a model with parametric memory alone.
03
When to use which
Use search when you need the exact wording of a clause, a statute or a figure.
Use an assistant to summarise, compare or draft, then open the sources it used.
Use an assistant with retrieval for questions such as “what did we agree with this client?”, where the answer lives in your own files.
Do not rely on an assistant’s memory for facts that must be right, such as case citations, dosages or tax rates.
04
Making an assistant behave more like good search
Show the retrieved passage and its file beside every answer.
Instruct the model to say “not in the documents” when it cannot find support, and test that it does.
Apply the same permissions to the assistant as to search, so staff see only what they may open.
Keep the index fresh when documents change.
Log the question, the sources used and the answer, so a doubtful result can be reconstructed.
05
Where ANVE fits
ANVE answers questions over the office’s own documents and shows the source passage, and its private GPTs are assistants built on the office’s own material. The research maker looks sources up online only when asked. ANVE is pre-release, and the software runs today on a development machine.
06
Questions people ask
Is an AI assistant a replacement for search?
Not for finding exact documents. Search shows the original; an assistant summarises it. Many offices need both, with the assistant grounded in the same index that search uses.
Why does an assistant sometimes cite a source that does not say what it claims?
Because the text and the citation are produced separately and can drift apart. In the 2023 audit only 74.5% of citations supported their associated sentence.
Can search be private too?
Yes. A search index over your own files can stay on your own systems. The privacy question is where any assistant on top of it runs.
Does retrieval make an assistant accurate?
It makes errors easier to check, not impossible: Stanford found 17% to 33% hallucination in retrieval-based legal tools.
07
What is not real yet
No sealed ANVE unit exists; every image of the appliance on this site is a labelled render. The software runs today on a development machine. No customers yet, no certifications, no delivery date, and the final price is not announced.
Reservations are ₹999 in India or US$12 elsewhere, refundable in full until you confirm an order in writing (terms). Offices that cannot use cloud AI can apply to be one of ten design partners.
