Gains:
- Ability to transform long industry reports, news and official data into summaries and insights structured with artificial intelligence and impose real resources on the model
- Ability to go beyond the raw summary by establishing the 'what does this mean' layer of synthesis with your own industry judgment
- Ability to rate source credibility and confirm each objective claim with at least two independent sources
One of the most labor-intensive stages of consultancy is research: reading dozens of pages of sector reports, annual activity reports, news, legislation and official statistics, and creating a meaningful picture from all of them. Artificial intelligence here reduces hours of reading-distilling work to minutes. But the real value of research synthesis is not the summary; "What does this mean?" is the answer to the question. In this unit, we will learn to use artificial intelligence as a reading and structuring engine, and to establish the interpretation layer of the synthesis with your own judgment.
What is synthesis, not summary?
An abstract is an abbreviated version of a text. Synthesis, on the other hand, is combining multiple sources and extracting new insight. “The report says the market grew by 8%” is a summary. "The 8% growth cited by all three sources comes largely from price growth; volume is almost constant, indicating that demand is saturated" is a synthesis. Artificial intelligence is very good at summary, but only you can direct it correctly and control the interpretation in synthesis.
Imposing the resource on the model
The most critical technique is this: telling the model to draw from the resource you give it, not from its own memory. Large language models do not "remember" up-to-date and accurate industry data; guesses. Therefore, pasting the actual report, news, or data into the text and saying "just bear with it" cuts off the hallucination to a large extent.
Your role: senior industry analyst for a management consultant. I have pasted 3 sources below (report, news, association data). Task: Create a structured summary for the Turkish [industry] market from these sources. Rules:- Rely ONLY on the sources I have given. Do not produce any numbers that are not in the source. - Write the source next to each number and claim: (Source 1, p. 4), etc. - If the sources contradict, state the contradiction clearly, do not make a choice. - At the end, add a list of "3 information that is not in the source but is necessary for the decision". Format: 1) Market size and growth 2) Driving forces 3) Risks 4) Uncertainties.
Tip: Leave a deliberate gap inside to test how well the model complies with the "rely on source only" instruction. For example, subtract the export figure from the sources; If the model still fits a number, don't trust that model/prompt pattern, tighten it up.
Dump long documents into structured output
Instead of crudely summarizing a 60-page report, ask directly for a decision table. Structured output is both faster to read and easier to verify.
Read the annual report below. Fill in the following table:| Topic | Finding in the report | Page | What it means to us (so-what) |Topics: revenue breakdown, geographic distribution, top risk, capital investment plan, management's expectation for next year. Leave the "What it means to us" column BLANK; I will fill it in. Mark the line where you cannot give a page reference as "no reference".
The critical detail in this prompt is this: you want the "so-what" column to be blank. The interpretation layer is the value added by the consultant; If you leave it to the model, it will produce a generic cliché.
Rating source credibility
Not all resources are equal. The power of a synthesis is as much as the power of the resources it is based on. Make rating below a habit.
Resource type
reliability
How to use
Audited financial statements
high
Direct quote, page reference
Official statistics (TURKSTAT, ministry)
high
Foundation for size and growth
Industry association report
Medium-High
For the trend; need cross checking
Serious press/analyst report
medium
Confirm with second source
Competitor marketing copy
Low-Medium
Claims may be exaggerated
Number from model's memory
low
Never count sources; verify
three mini cases
Case 1 — The price of trusting memory. A consultant asked "How big was the Turkish e-commerce market in 2024?" without citing the model. he asks; The model says "approximately 1.9 trillion TL". The consultant puts this in the presentation. The official report later reveals 2.8 trillion TL; A 30% deviation refutes the proposal's entire growth scenario. The correct way was to take the figure from the official source and just have it interpreted by the model.
Case 2 — Making contradiction visible. For a health technology project, two sources give very different market growth: the association report is 22%, the analyst rating is 11%. Because the consultant tells the model to "show the contradiction, don't select it," he presents both and finds that the difference is due to definition (one counts the device, the other counts the software). Instead of a single wrong number, the customer is presented with an accurate range and reason.
Case 3 — The value of synthesis. A consultant reads three separate sources: sales are increasing, the number of stores is stable, turnover per employee is increasing. The model summarizes these separately. The consultant establishes the synthesis himself: "Growth comes not from new stores, but from the efficiency of existing stores; so the growth strategy should not be 'more stores', but 'efficiency per store'." This is the insight that is not written in the source, but comes from the sources.
Weak prompt / Strong prompt
Weak prompt:
Summarize the Turkish cosmetics market.
It produces unsourced and possibly outdated text from model memory.
Powerful prompt:
Your role: industry analyst. I have pasted the 2024 cosmetics industry report and 3 news stories below.Task: Produce a structured synthesis from these sources only.- Write the source and page/URL next to each issue.- Separate based on the sources whether the growth comes from price or volume.- If the sources conflict, write the contradiction and the possible reason.- Leave the "So-what" lines blank.Probe: list 3 critical data that are missing for the decision.
Triangulation: intersection of three sources
Relying on a single source is risky; Every source has bias. Triangulation is confirming the same information from three different types of sources (e.g. official statistics, industry association and independent analyst). If three sources point in the same direction, trust increases; If it diverges, it means there is uncertainty and it should be reported instead of hiding it. AI speeds up triangulation: it can put the values of the same number from three sources side by side and calculate the difference. But you decide which three sources are truly independent; because often three "different" sources actually repeat (one quotes the other) the same original data, creating a false illusion of confirmation. Distinguishing whether a source is "original" or "transmitted" is the basis of the credibility of the synthesis.
Below are 3 sources that give the same market size number.Task: Put the three in the table; Write the value, year, method and original/transmitted distinction. Mark which of the sources is based on the other (if not independent). Tell me how many independent sources remain.
Research synthesis workflow
1) Collect sources and rate them for reliability.2) Feed the raw text to the model; say "just rely on this".3) Request structured table/output; leave so-what blank.4) Trace each issue back to the source; mark unreferenced ones.5) Keep contradictions, do not close them.6) Write the so-what/synthesis layer yourself.7) Confirm critical numbers with a second independent source.
Common mistakes
- Relying on the model's memory. The model does not remember current market data, it predicts it; Give real source.
- Mistaking summary for synthesis. “The report says this” is not enough; The "what does this mean" layer is the consultant's job.
- Hiding the contradiction. If two sources are different, choosing one and discarding the other produces false precision; Show the difference and why.
- Treating competitor marketing copy as fact. Statements such as "We are industry leaders" are often without evidence; confirm.
- Relying on a single source. Every number that will be the basis for a decision must be verified by at least two independent sources.
Caution: If a market size or growth number will be the basis of a quote or investment decision, that number will not be considered final until confirmed by at least three independent sources. In consulting, one made-up number destroys the credibility of the entire research in one sentence.
In summary
In market and industry research, AI saves hours of reading, distillation and structuring. The safest method is to give the real sources to the model and say "just rely on this", trace each number back to the source, and make the contradictions visible rather than hidden. But the spirit of the synthesis (what is actually happening in this market) and the verification of each number is in your hands. The model is a good reader and constructor; It is the consultant's duty to be a good interpreter and verifier.
Application task
Choose an industry and collect at least three factual/representative sources (reports, news, association data). Rate each source on reliability. Produce a structured output using the powerful synthesis prompt; Fill in the so-what column yourself. Distinguish based on sources whether growth is coming from price or volume. In the last step, confirm the most critical number with a second independent source and mark the ones you cannot confirm as "not confirmed".
checklist
- [ ] I gave the model a real source, I did not trust its memory.
- [ ] I traced each issue back to the source; I marked the ones without references.
- [ ] I rated the source reliability.
- [ ] I did not hide the contradictions, I showed them because of them.
- [ ] I wrote the so-what/synthesis layer myself.
- [ ] I have confirmed the critical numbers with at least two independent sources.
- [ ] I have separated competitor marketing claims from fact.