How everything on this site gets researched, checked, and corrected
Methodology · Last reviewed September 29, 2026 · 9-minute read · By Tom Morgan (TM)
AI Personalization is written by one person with a lot of help from AI tools. That combination can produce careful work or confident nonsense. This page explains what keeps it on the careful side, shows the mistakes I’ve already had to fix, and says plainly where I can’t make promises.
The short version
- AI tools help me research, outline, and draft. They never get the last word.
- Every number, date, and quote is traced to its original source before I publish. If I can’t verify it, it doesn’t run.
- Sources are ranked. A regulator’s text outranks an analyst survey, which outranks a vendor’s blog post.
- Anything I built on purpose, such as a scenario or a framework, is labeled as built.
- Mistakes are fixed in the open. The public corrections log is below.
On this page
- Who is accountable, and where AI fits
- Where AI-assisted writing goes wrong, and what stops it
- How an article gets made
- How I rank sources
- How I handle numbers
- Labels you’ll see
- What I don’t do
- Keeping articles current
- Corrections, and the public log
- Older articles
- Independence and funding
- Check any article in 60 seconds
- Questions
Who is accountable, and where AI fits {#who}
The About page says it plainly: I use tools like Grok and Claude, and I’m not a world-class expert in everything I write about. I’m not going to dress that up. What I can offer instead of credentials is a process you can inspect, which is what the rest of this page is.
Who’s accountable: me, Tom Morgan. Every article carries my name, and you can reach me through the Contact page or on X at @rkmlm1976. If something is wrong, that’s on me, not the tools.
| The tools do | I do |
|---|---|
| Round up and condense information from public sources | Check every fact, number, and quote against its source |
| Suggest outlines and ways to organize an argument | Decide what the conclusion is, and whether the piece is worth publishing |
| Produce early drafts so the piece flows | Rewrite, cut, and clarify until it reads like me |
| Publish nothing I haven’t reviewed, and own whatever is wrong |
Expertise on this site means method, not title: where a claim comes from, how strong that source is, and what would change my mind. Read my sources, not just my conclusions.
Where AI-assisted writing goes wrong, and what stops it {#failure-modes}
AI-assisted writing has known failure modes. Naming them is more useful than saying “I fact-check.” Here is each one and the specific control I use against it.
| Failure mode | What it looks like | Control |
|---|---|---|
| Invented statistic | A precise, plausible figure with no real source behind it | Any number I can’t find in an original source is cut |
| Misattribution | A real finding credited to the wrong study, firm, or year | Each figure is traced to the document that actually contains it, not to whichever page repeated it |
| Blended sources | Two reports with different samples merged into one claim | Statistics from different populations are never combined, even when they sound alike |
| Stale data presented as current | A “2026” report whose data was gathered earlier | I give both the data-collection date and the publication date when they differ |
| Vendor claim as benchmark | One company’s result presented as the industry norm | Vendor results are labeled as claims about that vendor only |
| Confidence masking uncertainty | Smooth prose that hides a disagreement between sources | Where analyst estimates conflict, I show the spread instead of picking one |
How an article gets made {#process}
- Start from a real question. I look for questions a reader would actually have: what a headline number really measures, what a tool really costs, when an approach works and when it doesn’t. Topics fall into five areas: AI and customer engagement, AI technology and innovation, AI-powered marketing, digital transformation with AI, and e-commerce personalization.
- Read the original material first. Regulation text, the study itself, product documentation, the company’s own announcement. Summaries and roundups are for finding leads, not for citing.
- Draft with AI assistance. The tools help with structure and a first pass. Everything they produce is treated as unverified.
- Verify line by line. Each figure, date, quote, and attribution is traced back. I check when the data was collected as well as when the report was published, and who paid for the research. Anything I can’t confirm is cut or marked as unconfirmed.
- Publish with a date, and stay reachable. If a reader finds a problem, the corrections process applies.
How I rank sources {#sources}
Not all sources deserve the same trust, and they don’t answer the same kinds of question. Where an article shows a Primary, Analyst, or Vendor badge next to a source, it means exactly this.
| Tier | What counts | How I use it | Caveat |
|---|---|---|---|
| Primary | Legislation and regulator guidance, peer-reviewed papers, official documentation and changelogs, company filings and announcements | Basis for factual claims | Primary isn’t neutral. A vendor’s documentation is authoritative on what its product does, not on how well it works |
| Analyst | Research firms, consultancies, industry surveys | Adoption and market trends, always with sample size, method, and dates | Firms often disagree widely on the same market, so I show the spread |
| Vendor | Company blogs, case studies, press releases | Claims about that company’s own deployments | Never presented as an industry benchmark |
| Lead only | Aggregators, listicles, SEO blogs, social posts | Finding the primary source | Never cited as evidence. If a statistic exists only here, I don’t publish it |
How I handle numbers {#numbers}
Numbers do most of the persuading in this field, so they get the most suspicion. Before a figure appears, it has to survive five questions. I think of it as a working checklist, not a validated method.
- Who measured it? An independent researcher, an analyst firm, or the company selling the product?
- What exactly was measured? The sample, the population, and whether it’s what people said or what was observed.
- When? A report labeled with a given year often contains data gathered earlier.
- How? Survey, experiment, log data, or modeled estimate.
- Who benefits if it’s big? That doesn’t make it false. It changes how much scrutiny it needs.
Two rules follow. First, I don’t merge statistics that only sound alike: a survey of executives’ beliefs and a measurement of actual outcomes are different populations. Second, a single company’s impressive result describes that company. For a worked example, see what GANNI’s “60% revenue” number actually measures.
A real example of the process catching my own error. In Demand Sensing vs. Demand Forecasting, a claim about SAP had been attributed to the wrong source, a copy-paste attribution error. Tracing it back showed the accurate, dated sources were SAP’s own announcements from 2026. I re-sourced it there and logged the fix below. That’s the whole method in one small case: follow the claim back until you reach the document that actually says it.
Labels you’ll see {#labels}
| Label | What it means | What it does not mean |
|---|---|---|
| Primary / Analyst / Vendor | Where the source sits on the ladder above | A guarantee that the source is right |
| Illustrative | A composite or hypothetical scenario built to explain a mechanism | A real client, a real result, or a documented case |
| Framework or heuristic | My own working rule of thumb for organizing a decision | A validated or statistically tested model |
| Estimate | Output of a simplified calculation from inputs you provide | A forecast or a promise about your results |
| Last updated | The date I last re-checked sources and changed something of substance | A date that moves for cosmetic edits |
| Correction note | A visible record that a figure or conclusion changed, and why | An attempt to bury the change |
What I don’t do {#limits}
I don’t invent statistics, case studies, testimonials, expert quotes, or benchmarks. I don’t present one source as settled when others disagree. I don’t give legal, financial, or regulatory advice; when I write about privacy law or compliance, it’s explanation, not counsel. See the Terms and Conditions for the formal version.
Most articles here are desk research. They synthesize public sources; they aren’t lab experiments. When I run my own comparison or test, the article says what I did and what I couldn’t control for. If it doesn’t say, assume I didn’t.
Keeping articles current {#updates}
I won’t promise a fixed review schedule for every article. At this scale it would be a promise I couldn’t keep. What I commit to is re-checking a piece when something material changes:
- A regulation’s dates or enforcement timeline move
- A vendor changes pricing or a feature the article depends on
- An analyst firm publishes a new edition of a report I cited
- A reader flags a problem
The “last updated” date changes only when the substance does.
Corrections, and the public log {#corrections}
If something looks shaky, outdated, or wrong, tell me through the Contact page, on X at @rkmlm1976, or in a comment. What happens next:
- I check the claim against its original source.
- If it’s wrong, I fix it. If the fix changes a figure or a conclusion, I add a visible correction note and update the date.
- If I think the article is right, I reply with the source that says so.
I can’t promise a reply time. I can promise the claim gets checked.
Corrections log
A methodology page that never shows a correction is asking for blind trust. Here are real fixes, including embarrassing ones.
| Date | Article | What was wrong | What changed |
|---|---|---|---|
| Sept 2026 | Demand Sensing vs. Demand Forecasting | A vendor claim was attributed to the wrong source; another rested on an unverifiable vendor blog; the market-size figure came from a competing vendor’s blog; the article carried an author bio and a first-person client anecdote I couldn’t substantiate; the publication date was in the future | Claims re-sourced to dated primary announcements; market size sourced to the research firm directly, with a note that analyst estimates differ widely; the bio and anecdote removed; date corrected |
| Aug 2026 | The 2026 AI Maturity Model | Some chart positions could be read as coming from one dataset; a bar chart could be misread on a linear scale | Positions labeled illustrative; chart annotated; direct links to primary sources added |
| Aug 2026 | AI Workflow Integration: Augment, Don’t Replace | Dates were stale (April instead of August 2026); in-page navigation links were broken | Dates refreshed; source-reliability badges added; navigation fixed |
This log covers fixes I’ve made and recorded. It isn’t a claim that nothing else is wrong. That’s why the next section exists.
Older articles {#legacy}
Some articles, especially from the site’s first months in late 2025, were published before I tightened this process. They may not meet every standard on this page. If you find one that doesn’t, tell me; I’d rather fix it or retire it than leave it standing on a claim I can’t support.
Independence and funding {#funding}
The site earns money from advertising. Ad placement is automated and separate from editorial decisions, and no advertiser or vendor pays for a review, a ranking, or a mention. If an article ever contains affiliate links, it will say so near the top. Data handling on the site is covered in the Privacy Policy.
Check any article in 60 seconds {#checklist}
You don’t have to take my word for any of this. Run this on any page here, or on anyone else’s.
- Does each key number link to a named source with a date?
- Does the article say whether that source is primary, analyst, or vendor?
- Can you tell who measured the number and whom they measured?
- Are scenarios and frameworks labeled as illustrative or heuristic?
- Is there a visible last-updated date, and do the sources plausibly predate it?
- Does the article tell you what it can’t say?
If an article here fails one of these, tell me.
The free tools follow the same logic: they’re estimators built on stated inputs, not oracles. The Personalization Maturity Assessment is a self-check to spot gaps, not a benchmark against other companies.
Questions {#faq}
Is the content on this site written by AI? AI tools such as Grok and Claude help with gathering public information, suggesting structure, and drafting early versions. The facts, numbers, quotes, and conclusions are checked and decided by me before anything goes live.
Why use AI at all if it can make mistakes? It speeds up the slow parts of research and drafting. The trade-off is that every AI-assisted claim is treated as unverified until it’s traced to a source, which is why the verification step and the failure-mode controls exist.
Can I trust the statistics you publish? Each one is traced to its original source and published with who measured it, who was measured, and when. Errors are still possible, and the corrections log shows some I’ve caught. Anything you find that looks wrong will be checked and corrected.
Do you test the products and tools you write about? Most articles are desk research that synthesizes public sources. When I’ve run a hands-on test or comparison, the article says what I did and what I didn’t control for.
How accurate are the free tools and calculators? They’re simplified models that turn your inputs into an estimate. They aren’t forecasts, and they can’t predict your specific results. Try the Personalization ROI Calculator with a range of inputs, not one.
Do you have formal credentials in this field? I’m not claiming any. What I offer is a documented process, named accountability, and a public record of corrections.
Do advertisers or vendors influence what you write? No advertiser or vendor pays for coverage, rankings, or reviews. See Independence and funding.
How do I report an error? Use the Contact page, message me on X at @rkmlm1976, or leave a comment. I’ll check the claim against its source and fix it if it’s wrong.
Thanks for reading how this place actually works. Hope it helps you trust what you read here for the right reasons, and question it when you shouldn’t.
Talk soon, TM—Tom Morgan