A team with no research budget finds original data in records it already keeps: support tickets, sales calls, order histories and site analytics.
Count one thing carefully. Then publish the method beside the number, so an AI engine can quote it whole.
Key Points
- In 2023, CMSWire reported that fewer than 30% of content marketers generated their own original research, while every one used other people’s research stats.
- A 2021 PLoS ONE study of senior marketers in 263 organizations linked regularly measuring content performance, and using that data, to higher effectiveness.
Support tickets, order records and sales notes already hold the counts a small team can publish.
Most small teams already hold a dataset nobody else can publish. Noticing it is the hard part. In 2018, research consultant Clare McDermott of Ravn Research named analyses of owned data, analyses of third-party data, short-form surveys and simple polls as workable alternatives to a costly benchmark study. Her examples were modest. Orbit Media judged the websites of the top 50 marketing and advertising companies against 10 web design standards, and by 2018 that 2015 post had earned hundreds of backlinks.
McDermott’s warning points the other way: research projects most often fail on poor survey design or faulty analysis, not on cost. So the sections below start with where your data already sits, move to how careful the count has to be, and then turn to shaping a finding so ChatGPT or Perplexity can lift it whole.
Questions this article answers
This article answers:
- Where does a team with no research budget get original data?
- How rigorous does no-budget data need to be before anyone trusts it?
- How do you turn first-party data into content AI engines cite?
Where does a team with no research budget get original data?
Original data usually sits in systems you already run: support tickets, sales and order records, site analytics and the product itself. Reading them closely produces facts no competitor can copy.
Before you pay for anything, work through these four checks:
- List every system that records a customer action, such as your help desk, CRM, order history and web analytics.
- Pick one question a buyer asks that those records can answer.
- Pull a single, clearly dated period of records.
- Compare what the records show with what your own pages claim.
The common assumption is that original data means running a survey. It does not. In its guide to content marketing research, Dovetail recommends a dual approach: pair light polling of current and potential customers with behavioral data from existing sales and other analytics, so stated preferences and actual behavior sit side by side. Learn why original data is the gap.
Anonymized usage data works as well. In 2018, the Content Marketing Institute described how Oh My Green, a startup that stocks office kitchens, published its most popular flavored LaCroix water from its own records. The Kitchn picked the study up, and a modest dataset became press coverage for a small food company.
How rigorous does no-budget data need to be before anyone trusts it?
Rigorous enough that a stranger could repeat your count. You need a stated method, an honest sample description and a clear limit on what the numbers can prove.
Formal research sets a high bar. Lisa Murton Beets, research director at the Content Marketing Institute, wrote in 2024 that high-quality research requires an understanding of methodology, sampling, questionnaire design, project management, data analysis and reporting. She also advised checking secondary sources first, including a brand’s own marketing stats and sales data. One market research firm’s 2021 example shows what the formal route involves: 400 responses gave a 5% margin of error at the 95% confidence level, and only with a probabilistic sample.
Most small teams cannot draw a random sample. That is acceptable if you say so. Washington State University‘s survey guidance accepts non-probability samples for exploratory work when resources are limited, but warns that the results cannot be generalized to the whole population. The same guide calls following up with non-respondents highly recommended, because skipping it can bias results.
I would not let any of this stop you. Owned records sidestep much of the sampling problem, because a help desk log for a given month is not a sample of that month’s tickets; it is all of them. What it cannot do is speak for customers who never wrote in. Say that plainly in your method line, and the finding becomes something an editor, or an answer engine, can quote without guessing at its limits.
How do you turn first-party data into content AI engines cite?
Package each finding so it stands alone: one number, its source, its date and the method, placed directly under a question-form heading that matches what buyers ask AI assistants.
Answer engines lift sentences, not reports. A finding buried on page nine of a PDF rarely surfaces. AEO Content, the company I co-founded, has run 26,577 real AI-visibility audits, and I read that volume as a reminder that ChatGPT, Perplexity and Google AI Overviews meet your work one page at a time. So each page needs a complete finding of its own:
- Put the headline number and its source in the same sentence.
- State the period covered and how you counted, in one line.
- Give each buyer question its own page or section.
- Repeat the count on a fixed schedule so the number stays current.
The fourth step is where small teams win. Michele Linn, former head of editorial at the Content Marketing Institute, said on The People of Digital Marketing podcast that CMI’s annual budgets, benchmarks and trends survey, first run in 2010, was its best-performing content year over year while she was there. A repeated benchmark compounds. Last year’s figure gives this year’s figure a story.
Speed is not the barrier people expect, either. One of the company’s client cases went from a quiet site to 10 booked calls in 3 weeks, and research-driven AEO content built on counts a team already holds is how that kind of result starts. The next decision is which buyer question your count answers first.
What should you do with your records this quarter?
Pick one record set you already own, count one thing in it, and publish the method beside the number. Owned sources keep claims checkable in a way borrowed ones do not.
The opening is wide. In 2023, CMSWire reported that fewer than 30% of content marketers generated their own original research, while every one of them used other people’s research stats. That imbalance is your advantage. Every borrowed figure in your niche already sits on many other pages, so ChatGPT or Google AI Overviews has no particular reason to cite your copy of it over the original. A count from your own support queue has only one possible source.
So open the ticket export before the survey tool. Tag a month of questions by topic, note how you tagged them, and put the most common one in a heading.
Frequently Asked Questions
What else do people ask about original data on a small budget?
Do I need to run a survey to have original data?
No. Washington State University’s survey guidance advises checking first whether secondary data analysis, meaning a fresh look at data that already exists, or interviews and focus groups can answer the question. A survey earns its place only when current data cannot.
Are surveys better than sales or support records?
They do different jobs. A survey tells you what people say they think, while order histories and ticket logs show what they actually did. I would pair the two before trusting either alone.
Can I add first-party data to an AI-assisted draft after it is written?
Yes, and I think that is the most practical place to start. Find each borrowed claim in the draft, replace it with a count from your own records, and write one sentence on how you counted it.
Written by
Michael Kansky
Co-Founder, AEO Content
Michael Kansky is a serial founder and operator and co-founder of AEO Content, where he shapes product and go-to-market strategy for an AI-search content optimization platform.
