AI & technology, made useful

Useful AI. Thoughtful tools. Better work.

Clear, human-first guides to AI utilities, digital workflows, privacy, and everyday technology.

Monday, November 2, 2026

Extract Data from PDFs with AI: A Review-First Workflow

A PDF table can look tidy on a screen and still turn into scrambled columns when copied elsewhere. AI and optical character recognition (OCR) can help reshape the information, but numbers, dates, and row boundaries need a deliberate review before you rely on them.

Check what kind of PDF you have

First try selecting a few words. If the text highlights cleanly, the document probably contains selectable text. If you can only select the whole page as an image, OCR is needed to recognize the characters. Scans often confuse similar marks such as 0 and O, 1 and I, decimal points, and faint minus signs.

Look at a representative page before processing a large file. A table that crosses a page break, uses merged headings, or has footnotes may need different handling from a simple list. Keep a copy of the original so you can compare every extracted value against its source.

Define the output before extraction

Decide what one row should represent and name the fields you need. For an invoice table, that might be item code, description, quantity, unit price, line amount, and source page. If a field is absent, ask the tool to leave it blank or mark it for review instead of filling it from a guess.

Keep source labels and units. “12” could mean pieces, dollars, or a page number; a bare value loses context. If the table is long, process a few pages at a time and include the page range in your request.

Preserve uncertainty instead of cleaning it away

Do not ask for a neat table at any cost. Tell the tool to mark unclear cells, retain the printed text when it is uncertain, and show which page each row came from. This gives you a review queue. You can then inspect just the uncertain entries in the PDF rather than checking every character from scratch.

A good first pass keeps source wording intact. Normalize dates, abbreviations, or currency only after you have confirmed the source format and chosen a rule. If different pages use different headings, preserve that variation in a note.

Verify the result before reuse

  • Compare the first and last row on each processed page with the PDF.
  • Check a few values from the middle, including decimals and negative values.
  • Recalculate visible totals and compare them with the printed total.
  • Confirm that rows were not merged or split at page breaks.
  • Keep the page reference until the extracted data has been approved.

For sensitive documents, follow your organization’s handling rules and use an approved tool. You can often test the workflow with a short synthetic table before uploading a real file. Extraction saves retyping; it does not remove the need to check the source.

Copy-ready prompts and example results

Open an example to view the complete prompt and a sample result. Results are illustrative; tools may respond differently.

Extract a short invoice table and check its total

Complete prompt

Extract the table below into CSV with these columns: item_code, description, quantity, unit_price, line_amount, source_page. Keep the printed values. Do not infer missing values. After the CSV, add the sum of line_amount and say whether it matches the printed total. Mark anything uncertain as REVIEW.

OCR text from PDF page 2:
R-11 | Notebook | 3 | 4.50 | 13.50
R-12 | Envelope | 2 | 1.25 | 2.50
R-13 | Labels | 1 | 6.00 | 6.00
Printed total: 22.00
Source page for all rows: 2

Illustrative expected result

item_code,description,quantity,unit_price,line_amount,source_page
R-11,Notebook,3,4.50,13.50,2
R-12,Envelope,2,1.25,2.50,2
R-13,Labels,1,6.00,6.00,2

Line amount sum: 13.50 + 2.50 + 6.00 = 22.00. It matches the printed total. No uncertainty is visible in the supplied OCR text.
Keep an unclear OCR value visible for review

Complete prompt

Convert these OCR lines into a table with columns date, reference, amount, and source_page. Do not correct a value by guessing. If a character is ambiguous, preserve the OCR text in raw_value and put REVIEW in status. Add a short note explaining what to inspect in the original PDF.

OCR text from page 4:
2026-03-08 | INV-2O4 | 18.50
2026-03-09 | INV-205 | 7.?0

The second amount is faint in the scan. The character in INV-2O4 may be a zero or the letter O.

Illustrative expected result

| date | reference | amount | source_page | status | note |
| 2026-03-08 | INV-2O4 | 18.50 | 4 | REVIEW | Check whether the third character in the reference is zero or letter O. |
| 2026-03-09 | INV-205 | 7.?0 | 4 | REVIEW | Inspect the faint amount in the original scan before entering a value. |

No ambiguous character has been silently corrected.

A Practical AI Workflow for a One-Person Content Studio

A solo creator can use AI to reduce blank-page work and organize repetitive steps. The distinctive value still comes from the creator’s reporting, judgment, experience, and editing.

Start with an original brief

Write the reader, question, promise, and scope in your own words. Note what you already know, what you need to verify, and what the piece will not claim. Ask an assistant to suggest missing questions or a possible outline, then revise it so the structure reflects the argument you actually want to make.

Do not let a generated outline become evidence. Research important claims using sources you can inspect. Keep links and notes beside the claims they support, and record when the information was checked if it can change over time.

Use AI for bounded production tasks

Once the research is in hand, use an assistant for narrow tasks: group interview notes, produce a rough structure, spot repeated ideas, or suggest alternate headings. Keep source material and unpublished client work private unless the tools and permissions you use allow it. If the assistant drafts prose, rewrite it to reflect your own reporting and voice, then check every factual statement.

For a content brief, you might ask: “Using only these notes, propose three possible section orders. Show the reader question each section answers. Do not add factual claims.” This gives you options without asking the tool to invent the substance.

Build a human review gate

  • Confirm the central claim and intended audience.
  • Verify numbers, quotations, dates, links, and product details.
  • Check whether the piece clearly distinguishes evidence from opinion.
  • Remove filler and language that does not sound like you.
  • Check image, quotation, and other usage permissions before publishing.
  • Label sponsored or affiliate relationships clearly and near the recommendation.

Keep the original research notes and a short change log. If a passage was substantially shaped by a tool, follow any applicable platform, publisher, client, or workplace disclosure rules. Do not imply that a tool was tested hands-on if it was not.

Repurpose with care

After the main piece is approved, ask for a short summary, a newsletter version, or a social caption based on the final article. Check each version against the published source. Context can disappear when a qualified statement becomes a short caption, so restore the nuance before sharing.

Review the workflow after several pieces. Keep the steps that save time without weakening quality, and remove any tool that produces more cleanup than value. A sustainable content process is measured by reader usefulness and editorial trust, not by how many drafts a system can generate.

Copy-ready prompts and example results

Open an example to view the complete prompt and a sample result. Results are illustrative; tools may respond differently.

Turn an original brief into a useful article outline

Complete prompt

Help me plan an original 700–900 word article for a one-person newsletter studio.

Topic: a five-minute pre-publish check for a weekly newsletter.
Audience: a solo creator who writes, edits, and publishes without a team.
Notes to use: check links; confirm dates and factual claims against their sources; confirm image permissions; preview on a phone; save a final copy.
Tone: calm, practical, and encouraging.
Format: a title, a short premise, five section headings with one-sentence notes, and a five-item checklist.
Limits: use only the notes above; do not promise that the checklist prevents every error. Flag any useful detail the brief does not supply.

Illustrative expected result

Title: A Five-Minute Newsletter Check Before You Publish
Premise: A compact routine catches common publishing issues while leaving final judgment with the creator.
1. Open every link — check that each destination matches its label.
2. Recheck dates and claims — compare them with the original notes or source.
3. Confirm image permission — record where the image came from and what use is allowed.
4. Preview on a phone — scan headings, spacing, and links at a small width.
5. Save the final version — keep the approved copy and publication date.

Checklist: links; dates and claims; image permissions; phone preview; final copy saved. Missing detail to confirm: the newsletter platform’s preview and archive steps.
Repurpose approved copy without adding claims

Complete prompt

Use only this approved source paragraph to create three small newsletter assets.

Source: “Each Thursday, the studio email shares two new pieces, a short note about how one piece was made, and one link to the shop. Subscribers can unsubscribe at any time.”

Create: (1) a social caption of at most 45 words, (2) three email subject lines of at most  eight words each, and (3) preview text of at most 80 characters. Keep the Thursday cadence and the contents accurate. Do not invent discounts, product materials, availability, or subscriber numbers. Label each item and include a final list of facts you preserved.

Illustrative expected result

Social caption: A small studio update lands every Thursday: two new pieces, a note from the making process, and a link to the shop. Join when it suits you; unsubscribe at any time.
Subject lines:
- Two new pieces this Thursday
- A note from the studio
- This week’s studio update
Preview text: Two new pieces, a making note, and a link to the shop.
Preserved facts: Thursday cadence; two pieces; one making note; shop link; unsubscribe option.

AI Assistant or Automation? How to Choose the Right Setup

A task that changes a little each time may benefit from an AI assistant. A task that follows the same clear rules may be better handled by ordinary automation. Many useful workflows combine both with a human approval step.

Separate variation from repetition

List the steps in the current process. Mark which steps are predictable and which require interpretation. Copying a value from one field to another is predictable. Deciding whether a vague message is a refund request may need context and judgment. Summarizing several different documents is variable, though the output format can still be fixed.

If the rules can be written as clear “when this, then that” conditions, start by considering a rule-based automation. It is usually easier to inspect why a rule ran and to test its edge cases. Use AI when the input is less structured and a draft or classification can be reviewed.

Choose the smallest useful level of autonomy

  • Assist: suggest, summarize, or draft while a person decides what to do.
  • Prepare: create a proposed record or message that waits for approval.
  • Act: make a change automatically under narrow, tested rules.

Start near the top of this list. Move toward automatic actions only after you understand the failure modes and have a way to undo or catch mistakes. For external messages, financial actions, permission changes, and important records, preserve a deliberate approval step unless a suitable control process is already in place.

Map a safe workflow

A practical pattern is: receive input, check whether it is complete, prepare a suggested result, show the evidence or reason, ask a person to approve, then record the outcome. Decide what happens when the tool is uncertain, unavailable, or wrong. A workflow with no safe fallback is not ready for unattended use.

Use a small set of representative examples before connecting real accounts. Test ordinary cases, missing information, duplicates, and unusual inputs. Keep access limited to the systems and actions the workflow needs, and revisit those permissions when the process changes.

Measure the whole process

Compare the time saved with setup, monitoring, exception handling, and correction time. Track errors and how quickly they are noticed. If an automation makes a small mistake repeatedly, the apparent time savings may disappear in cleanup.

The right design is the least complex one that handles the task reliably. Sometimes that is a checklist. Sometimes it is a simple rule. Sometimes AI helps prepare a draft. The useful question is which part of the work genuinely needs interpretation.

Copy-ready prompts and example results

Open an example to view the complete prompt and a sample result. Results are illustrative; tools may respond differently.

Use an assistant for uncertain support requests

Complete prompt

You are drafting triage notes for a small online stationery shop. Classify each message as FAQ, order action, or product problem. Assign urgency as routine, soon, or urgent using only the message text. Draft a reply only when a supplied store fact answers the question. Never send a message, change an order, promise a refund, or infer account details.

Return a table with: type, urgency, known fact, human next step, and optional reply. Store fact: Saturday hours are 10:00 to 14:00. No refund policy is supplied.

Messages:
1. Where can I find your Saturday opening hours?
2. My paper pad arrived soaked. Order PS-104.
3. Please cancel order PS-218; it has not shipped.
4. I cannot sign in and need to change the email on my account.

Illustrative expected result

| Message | Type | Urgency | Human next step | Optional reply |
| 1 | FAQ | Routine | None if the reply is approved | We are open Saturday from 10:00 to 14:00. |
| 2 | Product problem | Soon | Review the damage report and order PS-104 | I’m sorry the paper pad arrived damaged. Our team will review the order and follow up. |
| 3 | Order action | Soon | Check order status and confirm whether cancellation is possible | No cancellation promise; status must be checked first. |
| 4 | Account action | Soon | Use the approved identity-check process | No account change or identity assumption.
Automate fixed routing and stop at ambiguity

Complete prompt

Design a simple form-to-queue automation for a small support team. The form has one required field called Request type. Allowed values and destinations are: Billing goes to Finance; Login goes to Priority Support; General goes to Support. If the field is blank or contains any other value, send it to Manual Review.

The automation may assign a queue and add the received date. It must never approve refunds, change account access, or send a customer reply. Include the trigger, each rule, the fallback, one safety check, and a weekly measure of whether routing is working.

Illustrative expected result

Trigger: a new support form is submitted.
Rules: exact value Billing → Finance; Login → Priority Support; General → Support.
Fallback: blank or unrecognized value → Manual Review.
Allowed actions: assign queue and add received date only.
Safety check: test one sample for each allowed value plus blank and unrecognized values before turning on the workflow.
Weekly measure: count correctly routed forms and forms sent to Manual Review. Keep all customer decisions with a person.

How to Check and Fix Common AI Mistakes

AI output can be fluent and still be wrong. A quick review is enough for a low-stakes brainstorm; a factual or consequential answer needs checks matched to the possible harm.

Look for four common failure points

  • Invented detail: names, dates, features, or examples that were never provided.
  • Unsupported claim: a statement that sounds sourced but has no checkable evidence.
  • Hidden assumption: a missing condition quietly filled in with a guess.
  • Calculation or transformation error: arithmetic, units, copied rows, or formulas that do not match the inputs.

Also check whether the answer followed your requested scope. A response may be accurate in general and still fail the task by dropping a condition, changing the audience, or mixing old and new information.

Use a review sequence

First, compare the answer with the source material you gave it. Confirm quotations, names, quantities, dates, and any required points. Second, open external sources yourself and verify that each supports the linked claim. Third, reproduce calculations or test code and formulas on safe examples. Finally, inspect what is uncertain or missing.

For important work, get an independent check from a qualified person or an authoritative source. Asking the same AI system to review its own answer can surface issues, but it is not independent verification.

Ask for uncertainty in a useful form

Review the draft against the source text below.
Return a table with: statement | supported by source? | exact source passage | what remains uncertain.
Do not repair gaps by adding outside facts. Mark any unsupported statement for human review.
Draft and source: [paste the approved material]

This prompt can help organize a review, but you still need to inspect the cited passages. If the answer includes a confident statement without a source, treat it as a claim to check, not as evidence.

Match the review to the stakes

For a low-risk brainstorm, you may only need to remove irrelevant ideas. For a public article, confirm factual claims and usage rights. For decisions affecting health, money, employment, safety, or legal rights, rely on the relevant professional and primary records; do not let a generated answer make the decision for you.

Keep a brief note of recurring errors. If a tool repeatedly mishandles the same type of input, change the workflow, add a human check, or stop using it for that task. Reliable use includes knowing when the tool is not a good fit.

Copy-ready prompts and example results

Open an example to view the complete prompt and a sample result. Results are illustrative; tools may respond differently.

Check marketing copy against a short test record

Complete prompt

Compare the draft copy with the supplied test record. List each claim as supported, contradicted, or not established. Show the arithmetic for any percentage. Do not add product facts.

Test record: the current lamp uses 8 watts; the previous model used 10 watts under the same test; the bench test lasted two hours. No full-day test was run.

Draft copy: “Uses 40% less energy than the previous model and has been tested through a full working day.”

Return a table with claim, evidence, status, and a safer wording.

Illustrative expected result

| Claim | Evidence | Status | Safer wording |
| 40% less energy | 10 W − 8 W = 2 W; 2 ÷ 10 = 20% | Contradicted | “Used 20% less power than the previous model in the supplied test.” |
| Tested through a full working day | The record says the bench test lasted two hours | Contradicted | “Bench-tested for two hours.” |

The result applies only to the supplied comparison and test conditions.
Recalculate a summary from the numbers provided

Complete prompt

Audit this draft summary using only the table below. Recalculate the total, the percentage change from January to February, and the percentage change from February to March. Use the earlier month as the base for each percentage. Show the formula and round percentages to the nearest whole number.

Units sold: January 120; February 150; March 135.
Draft summary: “The quarter total was 420 units. Sales grew 20% from January to February, then fell 10% from February to March.”

Return the corrected figures and identify every error in the draft.

Illustrative expected result

Quarter total: 120 + 150 + 135 = 405 units.
January to February: (150 − 120) ÷ 120 = 25% increase.
February to March: (135 − 150) ÷ 150 = −10%, a 10% decrease.

The draft total and January-to-February percentage are incorrect. The February-to-March decrease is correct.

Compare AI Tools with a Fair, Repeatable Scorecard

A fair comparison starts with the work you need done. Feature lists and polished demonstrations can help you discover candidates, but only a repeatable trial shows how well a tool fits your own tasks.

Choose criteria before the trial

Pick a small set of criteria that matter to you. Useful options include output quality, time saved after checking, ease of use, privacy controls, accessibility, compatibility with your files, cost, and how easy it is to export your work. Give each criterion a simple weight from 1 to 3, where 3 means it matters most.

Do not let one impressive feature hide a serious problem. If privacy is essential, treat it as a minimum requirement rather than a score that can be outweighed by speed. Write down any “must pass” condition before comparing tools.

Use identical examples

Prepare three representative inputs: a common task, a more complex task, and an edge case. Use harmless or anonymized material. Give each tool the same instructions and the same time to complete the work. Keep a copy of the prompts and outputs so you can revisit the result instead of relying on memory.

Score each criterion from 1 to 5 and add one line of evidence. For example: “Quality 4 — kept all three required details after one edit.” A number without a reason is difficult to trust or reproduce.

Count the work that happens after generation

Measure the whole task: preparation, prompting, waiting, checking, editing, and moving the final result into your workflow. Note whether a tool made an error that would have been costly if missed. A quick response that needs extensive repair may be slower in practice than a less flashy alternative.

Check the current price and limits directly with the provider when you are ready to choose. Pricing and plan features change, so record the date and account type you evaluated. Include the time needed to learn the tool and the cost of keeping it connected to your workflow.

Make the comparison useful to someone else

Write down who the tool fits, who should skip it, the tasks you tried, the date of the check, and the limitations you found. Separate your observation (“it returned a table with two missing rows”) from your opinion (“that made review harder”). If a product has not been tested directly, label the information as an overview rather than a hands-on review.

The best scorecard is modest and transparent. It shows how the conclusion was reached and what would make you reach a different one.

Copy-ready prompts and example results

Open an example to view the complete prompt and a sample result. Results are illustrative; tools may respond differently.

Score two outputs against the same source notes

Complete prompt

Evaluate the two sample outputs using the same source notes and rubric. Score each criterion from 1 to 5, where 5 is strongest. Weight factual fidelity at 50%, preservation of open questions at 30%, and clarity at 20%. Explain any invented detail.

Source notes: Sam will send a revised quote by Wednesday. The installation date is not set. Check the voltage before booking.

Output Alder: “Sam will send the quote Wednesday. Installation is booked for Thursday. Check the voltage.”
Output Birch: “Sam will send the revised quote by Wednesday. The installation date is still open. Confirm the voltage before booking.”

Return criterion scores, weighted total out of 5, and a provisional winner. Treat this as a small sample, not a general product ranking.

Illustrative expected result

| Output | Factual fidelity | Open questions | Clarity | Weighted total |
| Alder | 3 | 2 | 4 | 2.9 / 5 |
| Birch | 5 | 5 | 5 | 5.0 / 5 |

Calculation for Alder: (3 × 0.50) + (2 × 0.30) + (4 × 0.20) = 2.9. The Thursday booking is invented and the open installation date is lost. Birch preserves the supplied details. Birch is the provisional winner for this one task only.
Calculate a weighted scorecard without hiding weak areas

Complete prompt

Calculate a weighted average using these fixed criteria: accuracy 50%, format fit 30%, and editing effort 20%. Multiply each 1–5 rating by its weight, then add the results. Do not change the weights.

Alder ratings: accuracy 4, format fit 5, editing effort 3.
Birch ratings: accuracy 3, format fit 4, editing effort 5.
Minimum accuracy requirement: 4 out of 5.

Return each weighted calculation, final score out of 5, whether it meets the minimum, and one caution about interpreting a short trial.

Illustrative expected result

Alder: (4 × 0.50) + (5 × 0.30) + (3 × 0.20) = 4.1 / 5. Meets the accuracy minimum.
Birch: (3 × 0.50) + (4 × 0.30) + (5 × 0.20) = 3.7 / 5. Does not meet the accuracy minimum.

Alder is the only candidate that meets the stated minimum in these ratings. Repeat the same tasks with more examples before making a broader claim.

Protect Your Privacy When Using AI Tools

Privacy is not a single toggle. It depends on what you share, which service receives it, how the service handles it, and whether you have permission to use that data there.

Classify the information before you paste

Pause before sharing account credentials, payment details, private messages, customer records, personal identifiers, unreleased work, or confidential documents. If the task does not require those details, remove them. Replace names with roles, round figures, and use a small synthetic example where possible.

For work information, follow your organization’s approved-tool list and data-handling rules. A personal preference setting does not override a contract, policy, or legal obligation. If you are unsure whether data may be shared, ask the data owner or administrator before proceeding.

Read the service settings that matter

Before relying on an AI service, review its current privacy notice and the controls in your account. Look for what the service says about submitted content, retention, model improvement, deletion, connected apps, and account access. These details can differ by product, account type, region, and plan, and they may change over time.

Do not assume that a private chat, a deleted conversation, or a paid plan means the same thing across services. Check the actual terms for the account you use. If an assistant can access files, email, or a calendar, review what it can read and what actions it can take.

Reduce exposure with practical habits

  • Share only the smallest excerpt needed for the task.
  • Remove direct identifiers and secrets from example inputs.
  • Turn off optional history or training controls if that fits your needs and the service offers them.
  • Use separate work and personal accounts where required.
  • Review connected apps and revoke access you no longer use.
  • Keep a human approval step before an assistant sends, publishes, or changes records.

When using a browser extension, phone app, or automation connected to an AI service, consider the whole path: what the extension can see, where prompts are sent, and where outputs are stored. A short privacy check is especially valuable when a workflow touches several services.

Make the decision proportional to the risk

A generic request to rewrite a public product description carries a different risk from a request involving medical, financial, employment, or customer data. Use stronger controls for sensitive tasks, and do not treat generated advice as a substitute for a qualified professional or an approved process.

Good privacy practice is usually simple: minimize, verify, and keep control. If the value of the task depends on sharing information you are not allowed to disclose, choose another method.

Copy-ready prompts and example results

Open an example to view the complete prompt and a sample result. Results are illustrative; tools may respond differently.

Redact a sample before asking for help

Complete prompt

Review the following synthetic customer note before it is used in a general-purpose AI tool. Do not repeat the name, email address, or order number in your answer.

Note: “Lila Sample, lila.sample@example.invalid, order DEMO-439: the package arrived three days late. Draft a brief apology.”

Return three items: (1) the kinds of data present, (2) a safer version with identifiers replaced by placeholders, and (3) one question I should check in my organization’s policy before using an external tool. Do not claim that a particular service stores or deletes data.

Illustrative expected result

Data present: direct contact details and an order reference, plus a service issue.
Safer version: “A customer says a package arrived three days late. Draft a brief apology.”
Policy check: confirm whether customer support information may be entered into the selected tool and account plan.
Compare supplied privacy statements without guessing

Complete prompt

Compare these fictional service summaries for a team that wants help rewriting non-confidential procedures.

Maple summary: personal chats may be used to improve the service unless a training control is switched off; chats remain until deleted by the user. The business plan says training is off by default and administrators can set retention.

Pine summary: personal chats are excluded from training; an abuse-prevention log may be kept for up to 30 days. Business retention options are described as contractual.

Create a table with training use, retention statement, and questions to verify. Use only the text above. Mark any item that needs confirmation in the current plan or contract. Do not treat these fictional summaries as real vendor terms.

Illustrative expected result

| Topic | Maple | Pine | Verify before use |
| Training | Personal plan may use chats unless the control is off; business summary says off by default | Personal chats are excluded according to the supplied summary | Confirm the setting for the exact account and plan |
| Retention | User deletion is mentioned; business administrators can set retention | Up to 30 days for an abuse-prevention log; business terms are contractual | Check the current contract, exceptions, and deletion process |

Neither summary alone establishes whether a particular internal procedure is approved for upload.

Use AI with Spreadsheets: A Safer Step-by-Step Method

AI can help explain a formula or suggest a transformation, but it may misunderstand column types, edge cases, and the way your workbook is used. Work on a copy and test the result before applying it to important data.

Describe the sheet, not just the desired answer

Share the column names, a few harmless sample rows, and the exact result you want. Explain how blanks, duplicates, dates, currencies, and unusual values should be handled. If the data identifies people, customers, employees, or clients, remove or replace those details unless your approved tools and policies allow the real data to be used.

For example, “standardize the date column” is ambiguous. “Convert values in column B to YYYY-MM-DD, leave blanks empty, and flag entries that cannot be parsed” is testable.

Request an explanation and a test case

Ask for the formula or steps, a plain-language explanation, and one example showing the input and expected output. If the answer contains a formula, confirm the target spreadsheet program and locale; function names and separators can differ. If it proposes code or a macro, read it and understand the effect before running it.

Given these column names and sample rows, suggest a formula for [specific result].
Explain each part and show the expected result for two sample rows.
State assumptions about blanks and text values. If the result depends on a missing detail, ask first.
Sample data: [anonymized columns and rows]

Test safely on a copy

  • Duplicate the sheet or save a version before editing.
  • Try the formula on a few known examples, including blank and unusual rows.
  • Compare the result with a manual calculation or an independent method.
  • Check totals, row counts, and any filters that could hide changed data.
  • Keep the original until you have reviewed the full result.

A plausible formula is not proof. Look for off-by-one ranges, text treated as numbers, dates interpreted differently, and formulas that skip hidden or blank cells. Test at the scale where an error would become visible without risking the original workbook.

Use AI as a spreadsheet tutor

If you already have a formula, ask for an explanation in ordinary language. Then check the explanation against the actual cell references. For a repetitive cleanup, ask for a documented sequence of steps rather than a one-click change. Keeping the method visible makes it easier for a colleague to review and repeat.

Copy-ready prompts and example results

Open an example to view the complete prompt and a sample result. Results are illustrative; tools may respond differently.

Ask for a formula and verify it on visible sample rows

Complete prompt

I use a spreadsheet where column B contains Category and column C contains Amount. Rows 2–5 are: Supplies, 120; Travel, 38; Travel, 265; Travel, 52.

Give me an Excel or Google Sheets formula to total only the Travel amounts in rows 2–5. Explain each range and show the expected result using these four rows. Do not modify the sheet or assume other data exists. Mention if a localized spreadsheet may require a different argument separator.

Illustrative expected result

Formula: =SUMIF(B2:B5,"Travel",C2:C5)
The first range checks the categories, “Travel” is the match, and the final range supplies the amounts to add.
Expected result: 38 + 265 + 52 = 355.
Some spreadsheet locales use semicolons instead of commas between formula arguments.
Flag duplicate invoice IDs on a copy of the sheet

Complete prompt

On a copy of my spreadsheet, column A contains Invoice ID in rows 2–6: INV-201, INV-204, INV-204, INV-205, INV-206.

Give me an Excel or Google Sheets formula for cell B2 that returns TRUE when the ID in that row appears more than once in A2:A6. Explain how to fill it down and show the expected TRUE/FALSE results for all five rows. Do not delete or merge records; a duplicate flag is only a review aid.

Illustrative expected result

Formula for B2: =COUNTIF($A$2:$A$6,A2)>1
Fill the formula from B2 down through B6. The dollar signs keep the full lookup range fixed while the row reference changes.
Expected results: INV-201 → FALSE; first INV-204 → TRUE; second INV-204 → TRUE; INV-205 → FALSE; INV-206 → FALSE.
Review flagged rows before changing any data.

Featured Post

Extract Data from PDFs with AI: A Review-First Workflow

Popular Posts