Original Research for SEO: How Tech Companies Earn Citations With Their Own Data

Original research for SEO is data only your company can produce, such as a survey, a benchmark or an analysis of your own records, published so a reader or an AI answer can cite you as the source.

An owner types a plain question into ChatGPT: how long does it take an MSP to onboard a 40-seat client? Back comes a number, and I’d like to know whose it is, because it won’t be yours. It probably traces to a vendor survey or an analyst note that one blogger summarized and another blogger summarized again.

Meanwhile the real onboarding dates for the real clients you’ve signed, every one of them timestamped in your PSA, sit unpublished, so nobody can quote them. A page built from borrowed percentages gives the engine and the buyer no reason to prefer it over the next page built from the same percentages.

Original research is the fix, and it’s less work than the word implies. Skip the thousand-person survey. Pick one question buyers keep asking, add records only you hold, and start there.

What Counts as Original Research, and What Doesn’t

Original research means you gathered the information yourself. A survey you ran, a benchmark you built, a pass through your own records, a test, interviews with named people. You can describe how you did it, and nobody else could have done it for you. Could a competitor rebuild the page with an afternoon of googling? Then it’s not research.

Google’s helpful content guidance asks whether a page provides “original information, reporting, research, or analysis.” It defines originality as “unique, original information or perspectives that aren’t already available on other websites.” And one line in there should be framed on a marketing office wall: “Attribution or giving credit to other sources doesn’t replace the need for original effort.”

That one stings. Twelve borrowed statistics with twelve tidy citations is a nicely formatted page. It isn’t original.

Original Research or a Roundup?

A roundup isn’t worthless. It’s a decent reference. It just isn’t research, and it shouldn’t be sold internally as the data asset.

What Google Promises, and What the Study Measured

Nothing. Start there.

Google’s May 21, 2025 Search Central post on AI search tells publishers to focus on “unique, non-commodity content that visitors from Search and your own readers will find helpful and satisfying.” Its page on AI features and your website says there are no additional requirements to appear in AI Overviews or AI Mode, and that no other special optimizations are necessary. Original research is the kind of page Google says it wants more of. That’s the whole claim. There’s no payout attached.

The research on answer engines is thinner than the blog posts about it. The paper behind the term GEO, Generative Engine Optimization by Aggarwal and five co-authors, was accepted to KDD 2024. It tested nine ways of raising a source’s visibility inside generated answers, and adding statistics, adding quotations from credible sources and citing sources led the field, with gains of up to 40 percent on a 10,000-query benchmark and up to 37 percent on Perplexity.ai. Keyword stuffing didn’t perform. The authors also say the results vary by domain.

Here’s what the paper doesn’t say. It rewrote existing pages to add statistics and quotes, so it didn’t test original research as such. It measured how prominently a source appeared in generated answers, not Google rankings, traffic or leads. Read it as evidence that answer engines surfaced pages with checkable numbers and named sources, and no further than that. Nothing more.

You’ll also run into vendor posts that quote multipliers for backlinks and AI citations. We couldn’t trace those figures to a published method. So we left them out. For the related question of whether AI-written pages get rewarded, see our post on whether AI content ranks.

Where an MSP’s Original Research Already Lives

Every MSP sits on a data set nobody else has. Your PSA, in a tool like ConnectWise PSA, Autotask or HaloPSA, holds every ticket you’ve closed. Your RMM holds every alert. Your finance records hold renewals, churn and what each client really costs to serve. Your quarterly business reviews hold what clients said they wanted, in their words. Nobody else has any of it.

What follows are questions, not findings. We haven’t run them, and no number in this post comes from one.

• Which ticket categories take longest to close for a client with 25 seats, and which for a client with 150?
• How long does onboarding really take, counted from signature to the last device enrolled?
• What do Microsoft 365 tenants look like the day you inherit them, before you’ve touched a setting?
• Which contract terms sit in front of renewals, and which sit in front of goodbyes?
• Win and loss notes, if anyone kept them honestly.

Pick the one a buyer would raise on a first call. That’s your question. The rest can wait for next quarter.

Client data needs a rule before it needs a chart. Aggregate it. Strip anything that names a client or a person. Read your contracts for confidentiality language, and get written clearance from whoever signs for your company. If a figure could identify one client, it doesn’t ship, however good the finding is. Do that first.

Once the research is live, other sites need a reason to mention it. Our guide to link building for MSPs covers the earning side, and our post on off-site mentions covers why those references carry weight with AI answers.

How to Publish Original Research So It Gets Cited

Good data on a bad page is wasted. Six steps, in order. Nothing exotic.

1. Pick one question a buyer would ask, and make it the title of the page.
2. Pull the data, and write down the sample, the period and what’s missing from it.
3. Write a method paragraph in plain language that someone else could follow and repeat.
4. Put the findings in HTML text and a table. Not in a PDF, and not in a picture of a chart.
5. Date the page, put a named author on it, and link to it from the service pages it supports.
6. Put the re-run on the calendar before you publish the first edition.

Step four is where good data goes to die. A crawler reads text. It can’t read a number locked inside an image, and it can’t see past a form.

Step six matters more than it looks. Orbit Media, a web design agency, runs a blogging survey on a fixed schedule, and its 2026 write-up reports on 1,042 content marketers and looks back over twelve years of answers. The figures are self-reported and come from the agency’s own audience, so weigh them that way. The format is the part to copy: a fixed question, a stated method and a date on every edition.

One word on markup. Google’s Dataset documentation covers markup that helps datasets show up in Dataset Search, so use it if you publish the dataset itself as a file that others can download and reuse. For a blog post’s findings, skip it. Google’s AI features page says no special schema is needed to appear in AI Overviews or AI Mode.

When to Skip It, and What to Do Instead

Skip it when you have no buyer question, nothing you can defend, a contract that bars it, or nobody who’ll own the re-run, since the first edition is easy and the second is where teams quietly stop. A study nobody refreshes ages in public.

Do the smaller thing instead. Interview two or three named people, use their words unchanged, and add your own analysis of what they said. Label it as interviews, not research. Be honest about it. Pull a few third-party figures, name the source and the year, and say what they mean for an MSP of your size. That’s an honest page. It isn’t a data asset, and it shouldn’t be dressed up as one.

Borrowed numbers rent you someone else’s authority. Your own records build yours. Systems first, then the page. If you want a second opinion on which of your records could carry one, book a growth strategy call.