Lesson objective

Rewrite instructional and statistical claims so that every teaching recommendation is labelled as practice and every survey figure stays inside the population it describes.

Plain-language explanation

Why this matters now

Every instructor has heard confident claims about teaching: that people remember a fixed percentage of what they hear and a larger one of what they do, that each learner has a visual or auditory style, that a certain method is "proven" to work. These claims travel from slide deck to slide deck, and they sound authoritative.

This programme holds no education or learning-science research. None. That is a deliberate limit on what you may say, not an oversight. It means that in this programme you may never assert a finding about how adults learn, how much they retain, whether learning styles exist, or which teaching method is more effective. You may describe what you do, why the structure of a small business makes it sensible, and label it: Best practice or AI Coalition Network recommendation.

The same discipline applies to the numbers you teach about AI in small business. Two credible surveys exist, and both carry limits. The Federal Reserve Banks' Small Business Credit Survey is not a probability sample. Its methodology says so plainly: "Samples for the SBCS are not selected randomly; thus, the SBCS may be subject to biases that random sample surveys would not be subject to." Its figures cover employer firms, while nonemployer firms appear in a separate chartbook with no AI finding. The Census Bureau's Business Trends and Outlook Survey covers employer businesses only, excludes farms, is experimental, and asks about AI use "in producing goods or services".

An instructor who teaches a finding that does not exist, or a number outside its population, repeats the error once for every learner. In this programme those learners have no compliance function to catch it.

A question to sit with

Look at the last slide deck you used to teach anything. How many of its claims about learning, or about how many businesses do something, could you trace to a named source with its population and date?

What goes wrong

1. The borrowed learning claim. A percentage for retention, a learning-styles quiz, or a line saying active learning is "proven" to work appears on a slide with no source. In this programme, that claim cannot be made at all.

2. Practice dressed up as science. A sensible habit, such as testing a procedure on a fictional draft, is justified with "the evidence shows". It needs no such justification, and the borrowed authority damages trust when a sceptic asks for the study.

3. The number leaves its population. An employer-firm figure is applied to a sole proprietor, or a BTOS figure for all employer businesses is quoted as the rate for cafés.

4. The self-report becomes a result. "71% said its use led to increased productivity", said by AI-using SBCS respondents, becomes "AI makes businesses 71% more productive". Those are different claims, and the second is false to the source.

5. Two surveys become a trend. The SBCS 46% from late 2025 and a BTOS rate near 20% from 2026 are placed side by side as if AI use fell, or as if one survey is wrong. They asked different questions of different samples. Neither is a trend line.

The practical answer

Label every teaching claim. Use these labels aloud and on slides.

Label What it means Example
Best practice Widely done, sensible, required by nobody "Test the procedure card on a fictional draft before the owner leaves."
AI Coalition Network recommendation This programme's educational standard "Every session ends with one written procedure per owner."
Survey evidence A figure from a named survey, with population, question, sample and date "In the 2025 SBCS of employer firms, a non-probability sample, 46% said..."
Your observation What you saw in your own sessions "In my sessions, owners rarely finish the card at home."

Your observation is legitimate and useful, as long as it is called an observation and never presented as a rate.

Rewrite the sentence, not just the slide. Most errors are in the sentence an instructor says without thinking. Practise the rewrites until they are automatic.

Tempting sentence Accurate sentence
"People remember most of what they do and little of what they hear." "I have owners write their own procedure because it gives them something to use on Monday. That is my practice, not a research finding."
"Half of small businesses use AI." "In the Federal Reserve's 2025 survey of employer firms, a non-probability sample, 46% said their business or its employees use AI."
"AI boosts productivity for 71% of small businesses." "Of employer firms in that survey that use AI, 71% said it increased productivity. That is what they reported, not a measured result."
"Businesses like yours, with no staff, mostly use AI now." "I don't have a reliable figure for businesses with no employees."
"AI use dropped from 46% to 20%." "Two surveys asked different questions of different samples, so their figures differ. Neither shows a trend."

Know the four facts that keep a figure honest. For the SBCS employer report: 6,525 responses; firms with 1 to 499 employees; fielded 3 September to 14 November 2025; a non-probability sample weighted by raking to match the small firm population across firm and owner characteristics. For the BTOS story: employer businesses, farms excluded, an experimental product, a question about producing goods or services, and a period from 14 December 2025 to 3 May 2026. If you cannot state the four facts, do not quote the figure.

Best practice: keep a claims list for each session. Before delivery, list every factual claim on your slides and in your script, with its source key or "practice" or "observation" beside it. Anything with no entry is cut. Module 03 turns this list into a currency log with a date and a checker.

Answer the sceptic with the method, not the number. When an owner says "so nobody knows anything?", the honest answer is that the surveys tell us some things about employer firms, with limits, and that the way to know about your own business is to measure your own work. That is a stronger position than any borrowed figure.

Remember this

  • This programme holds no learning-science research; never assert a finding about learning, retention, styles or method effectiveness.
  • Label every teaching claim: best practice, AI Coalition Network recommendation, survey evidence, or your own observation.
  • Every survey figure stays inside its population, question, sample and date.
  • Self-reported outcomes are what respondents said, not measured results.
  • Keep a claims list and cut any claim with no source or label.

Small business example

Imani Brooke, a fictional community college instructor, is asked to review a colleague's slide deck before a fictional evening course for small business owners. The deck has twenty-four slides. She makes a claims list and finds eleven factual claims. Three are about learning: a slide giving a precise percentage for how much people remember of what they hear compared with what they do, a slide asking learners to identify their "learning style", and a sentence calling hands-on practice "proven" to be twice as effective. None has a source, and this programme holds no learning-science research, so all three are cut or rewritten as practice: "We will spend most of tonight writing and testing your own procedure. That is how I run these sessions; it is my practice, not a research finding." Four claims are survey figures. One says "46% of small businesses use AI" with no population; she rewrites it with its employer-firm population, its question and the non-probability caveat. One says "AI raised productivity at 71% of firms"; she rewrites it as what AI-using employer respondents said. One says "most sole traders now use AI"; there is no verified figure, so she replaces it with the sentence "I don't have a reliable figure for businesses with no employees." One says "AI use among small firms fell from 46% to 20% this year"; she replaces it with a short explanation that the two surveys asked different questions. The remaining four claims are the colleague's own observations from past sessions, which she keeps with the label "in my sessions". Eleven claims: three cut or rewritten as practice, four survey figures corrected, four observations kept. She returns the list with each claim's label and source key.

Responsible-use warning

Never present a retention percentage, a learning-styles claim or a teaching-method comparison as a finding, even when it appears in someone else's respected deck. Never attach a claimed study to a practice that is sensible on its own terms. Never quote a survey figure without its population, question and sampling caveat, never quote a precise BTOS figure for the smallest firms, and never present the SBCS and BTOS figures as a trend or a contradiction.

Human-review requirement

The instructor owns every claim in the deck, including any drafted with an AI tool. Before delivery, a second reviewer reads the claims list line by line, confirms that each learning claim is labelled as practice or observation, that each survey figure carries its population, question, sampling caveat and date, and that no employer figure is applied to nonemployer firms. The reviewer records the date of the check and their name, so that the claims list can become the session's currency log.

Practical exercise

Audit a fictional slide deck for borrowed findings and borrowed numbers. Use a fictional deck and fictional session throughout.

  1. Write a fictional ten-slide deck for a small business AI session that contains at least two claims about learning, three survey figures and two of the instructor's own observations.
  2. Build a claims list: number every factual claim and write its text exactly as it appears.
  3. Beside each claim write its label: best practice, AI Coalition Network recommendation, survey evidence, or your observation. Mark any claim that fits none of these.
  4. For each learning claim, rewrite it as a practice statement or delete it, and record which you did.
  5. For each survey figure, write its population, the question asked, whether the sample was random, and the period covered. Rewrite any figure missing one of these.
  6. Mark any figure applied to businesses with no employees and replace it with the honest sentence.
  7. Count your claims by label and confirm the counts add up to the total on your list.
  8. Date the list, sign it as reviewer, and keep it to become the session's currency log in Module 03, where each claim gains a last-checked date and the name of the person who checked it.

Knowledge check

  1. May an instructor in this programme state a retention percentage if it appears in a respected source they have seen elsewhere?

    Answer guidance: No. This programme holds no learning-science research, so no finding about learning, retention, learning styles or method effectiveness may be asserted; the practice may be described and labelled as practice.

  2. What four labels may a teaching claim carry?

    Answer guidance: Best practice, AI Coalition Network recommendation, survey evidence with its population and caveats, or the instructor's own observation, called an observation and never a rate.

  3. Why should the SBCS 46% and the BTOS figure near 20% never be presented as a fall in AI use?

    Answer guidance: Because they asked different questions of different samples: the SBCS counted use by the business or its employees in a non-probability sample, while BTOS asked about producing goods or services.

  4. What is a claims list and what happens to a claim with no entry beside it?

    Answer guidance: A list of every factual claim in the session with its source key or label beside it; any claim that has no source and no label is cut before delivery.

Sources and further reading