---
title: "AEO/GEO — Practical workbook"
language: en-CA
sourceKey: "aeo-geo-series-2026-10-22:public-workbook"
reviewNotes: "Workbook revised at the owner’s request on 2026-10-26 for consistency with the ten articles, trial budget and reviews. Worksheets A–I and fictional examples have equivalent French and English scope."
---

# AEO/GEO — Practical workbook

**TechnoConception — A ten-part series**

[Start with episode 1: Your customers may compare you before visiting your website](/insights/aeo-geo-before-first-visit/)

## Practical workbook — Bilingual panel, observations, and CRM

This proposed sample should be adapted to your buyers’ questions. It uses a fictional scenario: a Quebec service business is evaluating a CRM and the support needed to implement it. It does not present the results of a completed audit or describe a TechnoConception customer or offering.

### 1. A panel of twelve buying questions

Keep the questions and context constant throughout a series of observations. Analyse languages and intentions separately. **Define the expected signal before each test**: provider recommendations for D1 and D2; an informative answer and its sources for D3, D4, and C1 to C4; an accurate representation of the provider for V1 to V4. An informational answer can be correct without mentioning the company being audited.

**French context to include with every question:** « Nous sommes une PME québécoise de services. Notre équipe commerciale compte six personnes et travaille en français et en anglais. Nous évaluons un CRM et le soutien nécessaire à son déploiement. »

**Equivalent English context:** “We are a Quebec service business. Our six-person sales team works in French and English. We are evaluating a CRM and the support needed to roll it out.”

The first eight questions are unbranded. The last four are **branded templates**: replace `[nom réel du prestataire retenu]` and `[verified provider name]` with the same real provider name, verified and fixed before the audit. Do not run an unresolved template or ask the model to invent the provider. Do not insert that name into the unbranded questions.

| No. | Stage and type | French question | Natural English equivalent |
|---|---|---|---|
| D1 | Discovery · unbranded | Quels prestataires au Québec pourraient nous aider à choisir un CRM adapté à notre PME et à son équipe bilingue? Nomme des options pertinentes, cite leurs sources publiques et indique ce qui reste à vérifier. | Which providers in Quebec could help us choose a CRM suited to our small business and bilingual team? Name relevant options, cite their public sources, and identify what still needs verification. |
| D2 | Discovery · unbranded | Quels prestataires au Québec pourraient prendre en charge la migration de nos données, la configuration du CRM et la formation? Nomme-les, cite les services publiquement confirmés et précise les responsabilités inconnues. | Which providers in Quebec could handle our data migration, CRM configuration, and training? Name them, cite publicly confirmed services, and identify responsibilities that remain unknown. |
| D3 | Discovery · unbranded | Quelles questions poser avant de retenir un CRM et un accompagnement adaptés au français et à l’anglais? | What should we ask before choosing a CRM and rollout support that work in French and English? |
| D4 | Discovery · unbranded | Quelles informations préparer avant de demander une proposition pour remplacer nos fichiers clients par un CRM? | What information should we prepare before requesting a proposal to replace our customer spreadsheets with a CRM? |
| C1 | Comparison · unbranded | Comment comparer un déploiement réalisé à l’interne, un accompagnement et une réalisation confiée à un prestataire? | How should we compare an internal rollout, guided implementation, and a rollout delivered by a service provider? |
| C2 | Comparison · unbranded | Comment comparer une proposition CRM incluant migration et formation avec une proposition couvrant seulement la configuration? | How should we compare a CRM proposal including migration and training with one covering configuration only? |
| C3 | Comparison · unbranded | Comment comparer les coûts sur douze mois, y compris les licences, la préparation des données, le soutien et le temps de notre équipe? | How should we compare twelve-month costs, including licences, data preparation, support, and our team's time? |
| C4 | Comparison · unbranded | Comment évaluer un prestataire CRM qui insiste sur les fonctionnalités et un autre qui décrit surtout sa méthode et ses livrables? | How should we assess a CRM provider emphasizing features against one focusing on its process and deliverables? |
| V1 | Validation · provider name required | Quels livrables [nom réel du prestataire retenu] propose-t-il pour notre projet CRM? Cite les sources publiques et distingue les informations confirmées des questions ouvertes. | What deliverables does [verified provider name] offer for our CRM project? Cite public sources and separate confirmed information from open questions. |
| V2 | Validation · provider name required | Que peut-on vérifier publiquement sur la migration, la configuration et la formation proposées par [nom réel du prestataire retenu]? Cite les pages et signale les inconnues. | What can we publicly verify about migration, configuration, and training offered by [verified provider name]? Cite the pages and identify unknowns. |
| V3 | Validation · provider name required | Quelles preuves publiques permettent d’évaluer l’expérience de [nom réel du prestataire retenu] auprès de PME comparables? Distingue ses affirmations des témoignages et références indépendantes. | What public evidence helps assess [verified provider name]'s experience with similar businesses? Distinguish its claims from testimonials and independent references. |
| V4 | Validation · provider name required | Quels points faut-il clarifier avec [nom réel du prestataire retenu] sur les responsabilités, exclusions, soutien après déploiement et récupération des données? Appuie-toi sur les informations publiques et indique les questions ouvertes. | What should we clarify with [verified provider name] about responsibilities, exclusions, support after rollout, and retrieving our data? Use public information and identify open questions. |

#### Test size and scope

Proposed first pass: **12 questions × 1 interface × 1 language × 2 repetitions = 24 observations**. Expanded version: **12 questions × 4 interfaces × 2 languages × 3 repetitions = 288 observations**, provided the interfaces are accessible. This protocol is neither a standard nor a statistically validated sample size.

Planned unbranded and branded runs total **16 and 8**, respectively, in the first pass, and **192 and 96** in the expanded version. Within these groups, D1 and D2 explicitly call for provider recommendations: **4 observations** initially and **48** in the expanded version. Report each result against the question group that matches its expected signal, stating how many answers were actually generated and evaluated.

A branded answer measures how the provider is represented when its name is supplied. It does not measure spontaneous discovery. Repetitions record observed responses, not people or search volume. Keep unavailable conditions separate. The results describe this panel, not market share.

### 2. A reusable observation record

One row represents one attempt. Distinguish a generated AI answer, a valid search with no AI Overview, an unavailable feature, and a technical error. A search with no Overview counts towards the frequency with which Overviews appear; only generated answers enter the denominators for mention or citation rates within answers. A failed attempt remains recorded with its reason and does not count among the evaluated answers.

| Field | What to record |
|---|---|
| Observation | Identifier, question number, and repetition number. |
| Exact question | Submitted text, context, and the name actually inserted. |
| Type and expected signal | Stage; branded or unbranded; sourced information, recommendation, or accurate representation. |
| Interface and model | Product and surface used; displayed model or “not disclosed.” |
| Language and region | Language of the question and answer; stated region, then known technical location or setting. Preserve unknowns. |
| Date | Date, time, and time zone. |
| Conditions | Fresh conversation, or a new search for AI Overviews; search mode; known sign-in and personalization conditions. A fresh conversation does not prove personalization is absent. |
| Raw result | Answer and sources, valid search with no AI Overview, unavailability, or failure; no unnecessary customer data. |
| Mention | Brand present, absent, or undetermined; relevant sentence. |
| Citation | Exact URL and claim supported; provider source or third-party source. |
| Recommendation | Provider suggested as a suitable option; conditions and reservations. A mention alone is insufficient. |
| Accuracy | Claim confirmed, contradicted, unverifiable, or outdated; verification source and date. |
| Uncertainty | Missing information, disagreements, and verification limits. |
| Action | Correction, clarification, evidence to add, or technical investigation; owner. |

Keep these dimensions separate: a citation can be accurate without recommending the company, and a recommendation can contain an error.

### 3. Connect enquiries to the CRM

Keep technical records, the prospect’s statements, and business outcomes in separate fields.

| Group | Useful fields | Interpretation rule |
|---|---|---|
| Observed visit | Referrer, landing page, date, time, permitted identifier, and known campaign. | A missing referrer stays unknown; “direct” does not automatically become “AI.” |
| Self-reported discovery | Optional question: “How did you hear about us?”; answer and named tool. | Complements technical records without replacing them. |
| Need and fit | Need, scope, responsibilities, timeline, and agreed qualification criteria. | Define qualification before tracking; a completed form is not enough. |
| Sales opportunity | Status, next step, owner, documented estimated value, and reason for not proceeding. | Retain lost, unqualified, and unanswered enquiries too. |
| Outcome | Customer won or not; date; contract value or recognized revenue, clearly distinguished. | An estimated opportunity value is not realized revenue. |
| Matching records | Prospect or company identifier; associated enquiries and visits. | Deduplicate submissions; multiple contacts do not create multiple sales. |
| Attribution limits | Unknown source, unconfirmed statement, multiple channels, or records that cannot be matched. | Preserve these categories in the review. |

Limit collection to information useful for follow-up. The channel’s name is generally enough; there is no need to request the prospect’s entire conversation. Restrict data access to the people who need it.

### 4. A fictional numerical example

These invented numbers illustrate the calculation; they represent neither TechnoConception nor a study. Both arrival cohorts cover the same period. Every arrival has the same follow-up period, fully elapsed by the review date. Each unique enquiry and each customer won is attributed to **one channel only**, under a rule set in advance. The rates use **sessions**, not unique people, as their denominator.

| Measure | Visits attributed to an AI source | Organic search |
|---|---:|---:|
| Sessions | 100 | 1,000 |
| Enquiries received | 8 | 40 |
| Qualified enquiries | 3 | 12 |
| Customers won | 1 | 4 |
| Enquiries per session | 8 / 100 = **8%** | 40 / 1,000 = **4%** |
| Qualified enquiries per session | 3 / 100 = **3%** | 12 / 1,000 = **1.2%** |
| Customers won per session | 1 / 100 = **1%** | 4 / 1,000 = **0.4%** |
| Qualified enquiries among enquiries | 3 / 8 = **37.5%** | 12 / 40 = **30%** |

The AI cohort has twice the enquiry rate, but contributes three qualified enquiries and one customer, compared with twelve and four from organic search. Keep volumes, contract values, and costs alongside the rates.

One additional customer would take the AI rate from 1% to 2%. These small numbers do not support a stable forecast. The cohorts may differ in intent or landing pages. Attribution describes the available records and rules; it proves neither that AI caused the sale nor that an AEO/GEO action produced the difference.

## Complementary worksheets — From question to test

The following templates propose a working method. They accompany the ten articles; they do not represent a completed audit or an existing contractual offering. Their purpose is to make each action traceable and open to revision.

### A. A buying-question worksheet

Complete one record per need, then group formulations that call for the same decision and evidence.

| Field | What to enter | Entirely fictional example |
|---|---|---|
| Audience | Relevant role, size, and context | Manager of a service business with six salespeople |
| Question received | Actual wording and anonymized origin | “Who prepares our data before the CRM migration?” — sales call |
| Decision | The choice the answer should support | Preparing the data internally or assigning the work to a provider |
| Constraints | Language, service area, budget if known, dependencies | French/English, Quebec, export from two spreadsheets |
| Expected evidence | Information needed to make the answer credible | Scope, responsibilities, sample deliverable |
| Existing page | Current URL or confirmed absence | Selection-service page; migration still covered only briefly |
| Gap | Missing information or contradiction | The page does not specify who removes duplicate records |
| Proposed action | Correction or content to produce | Add a responsibility matrix validated by the team |
| Future observation | What will be checked after the action | Content access, accurate description in the panel, incoming buying questions |

Use three simple criteria to prioritize topics: proximity to a decision, seriousness of an incorrect answer, and ability to provide useful evidence. An ordinal score from 1 to 3 can help order a list; it is not a revenue forecast. Preserving the question’s origin helps avoid treating an assistant-generated suggestion as a measure of real demand.

**Suggested instruction for using an assistant with anonymized notes:**

> Using only the notes below, extract the buying questions actually present. For each one, identify the decision, constraints, the excerpt supporting its inclusion, and missing information. Group equivalent formulations. Place any new idea in a separate section titled “hypotheses to validate.” Do not invent search volume, customers, or product capabilities. Propose French and English wording that preserves the same need.

This is a template for notes you are authorized to use; it has not been run on customer data for this workbook.

### B. A technical correction ticket

| Item | Expected content |
|---|---|
| Business objective | Intended search use and relevant public content; separate decisions about model training. |
| Observation | URL, date, tool or crawler concerned, observed response, and text actually received. |
| Verified cause or hypothesis | Identified blocking rule, confirmed rendering problem, or investigation still open. |
| Intervention | Specific setting to change and qualified owner. |
| Validation | Expected content received under the agreed conditions; retain the evidence. |
| Rollback | Previous configuration and how to restore it. |
| Later check | New observation in the panel; distinguish technical success from citation. |

Avoid a ticket that says “unblock all AI.” A useful ticket names the service, use, page, and intended technical outcome. A major architecture change should address an observed obstacle; it does not follow solely from a wish to improve a visibility score.

### C. An offer page your team can substantiate

Before approving a page, have an operational owner answer these questions:

1. Who can benefit from this offering, and in what context?
2. Which deliverables are actually provided?
3. Who performs each part of the work, especially data preparation, configuration, and training?
4. Which dependencies, exclusions, or limits could change the decision?
5. What public evidence or demonstration supports each important claim?
6. What date, version, or condition makes that evidence interpretable?
7. What next step can readers take without assuming a capability that does not exist?

A useful comparison uses the same criteria for each option, a verification date, and the corresponding sources. “Unknown” is an acceptable value. One tool’s starting price should not be compared with another’s full cost without explanation. A fictional example stays labelled; a customer case study requires suitable data and permission.

### D. An external-source and contribution log

| Field | Purpose |
|---|---|
| Source and exact URL | Identify the page or thread, not just the domain. |
| Relevant question | Connect the action to an observed buyer need. |
| Category | Managed listing, independent editorial content, sponsorship, review, community, or company-owned content. |
| Claim to verify | Record the exact passage and the reference information. |
| Claim origin | Distinguish original research, an interview, and a republished press release. |
| Proposed contribution | Factual correction, evidence, demonstration, expert opinion, or sponsored format. |
| Publishing decision-maker | Identify who can decide; no promise of editorial placement. |
| Cost and deliverable | Describe what is being purchased, if applicable, without promising an AI recommendation. |
| Status and outcome | Prepared, sent, accepted, published, rejected; date and verified link. |
| Follow-up | Access, accuracy, observed presence, and enquiries; keep each measure separate. |

Do not treat ten republications of the same text as ten independent testimonials. A correction can be useful without a new mention: its value may lie in preventing an enquiry outside the business’s scope.

### E. An expert interview and reuse plan

**Proposed format: 30 minutes, adjustable to the topic.** Prepare questions and documents before the interview; obtain the person’s agreement to the intended uses.

| Suggested time | Question | Material sought |
|---|---|---|
| 0–5 minutes | What decision is your customer actually trying to make? | Audience, problem, and consequences |
| 5–12 minutes | Can you show a specific example? | Demonstration, anonymized deliverable, steps |
| 12–18 minutes | When is this method unsuitable? | Limits, dependencies, alternatives |
| 18–24 minutes | What mistake keeps recurring? How can someone recognize it? | Verifiable criteria and a useful caution |
| 24–30 minutes | What should the customer prepare to move forward? | Action, documents, and responsibilities |

Use the interview to produce a video focused on one question, an article that places the evidence in context, a short answer to an objection, and a social excerpt that retains the essential qualification. Check names, numbers, negations, and limits in the transcript. The versions should remain consistent even when they differ in length.

**Suggested transformation instruction:**

> Turn this authorized transcript into an article draft. Preserve conditions and uncertainties. For each important factual claim, identify the source passage and any external evidence that may be needed. Separate authorized real examples, opinions, and fictional illustrations. Do not turn any estimate into a measured result. Finish with the questions the expert must check before publication.

### F. An experiment record and decision rule

| Item | Example structure to complete |
|---|---|
| Hypothesis | A specific clarification will better answer a defined buying question. |
| Intervention | One documented family of changes; pages and date. |
| Treated group | Set of affected topics/pages, defined before the intervention. |
| Comparison | Topics/pages as comparable as possible, left unchanged during the period. |
| Possible contamination | Shared pages, shared brand, internal links, a campaign, or an external publication. |
| Primary measure | For example, proportion of accurate answers for the defined need. |
| Other measures | Relevant citations, visits, qualified enquiries; the denominator for each. |
| Conditions | Panel version, languages, surfaces, repetitions, settings, and handling of failures. |
| Duration | Baseline and follow-up periods suited to the observation frequency and sales cycle. |
| Cost | Internal time, provider, tool, production, and distribution. |
| Planned decision | Continue, revise, extend observation, or stop under criteria set in advance. |

**Entirely fictional illustrative calculation:** accurate answers increase from 20% to 35% in the treated group and from 25% to 30% in the comparison group. The descriptive contrast is (35−20)−(30−25) = **10 percentage points**, not a demonstrated 10% increase in sales. Retain the counts, differences between topics, variation over time, and platform changes. This calculation alone does not prove a causal effect. A business with few topics may use a gradual rollout and remain more cautious in its conclusion.

### G. Evaluate a tracking tool before committing

Request a demonstration on a small, relevant panel, with the conditions documented. Compare tools by what you will be able to understand and export, not just the number of model logos.

| Question for the supplier | Expected usable answer |
|---|---|
| Where do the questions come from? | Clearly distinguish selected questions, observed data, and estimates. |
| How do you obtain the answers? | API or interface, search mode, region, and known limitations. |
| How is the score calculated? | Numerator, denominator, weighting, and excluded cases. |
| What happens after a failure? | Visible failure, retry rules, and history. |
| Can we keep a constant panel? | Versions, change history, and comparisons with the same scope. |
| Can we export answers and URLs? | Usable raw data with dates and settings. |
| How do you count the brand? | Namesakes, variations, parent/subsidiary relationships, and disambiguation. |
| How do you check accuracy and sentiment? | Criteria, examples, and an option for human review. |
| Which costs change with usage? | Clearly priced prompts, surfaces, languages, repetitions, users, and exports. |
| What remains after cancellation? | Export, historical access, and obligations described in the contract. |

The same tool may suit one business and be excessive for another. An API-based measure can support consistent tracking; present it as that specific measure. A capture from the user interface may more closely resemble an observed journey without becoming representative of all users. Neither method, on its own, reveals every question the market actually asks.

### H. Budget and reviews for the 90-day trial

Define one offer, the pages involved and the available time before choosing the panel size. The illustrative 24-hour limit in part 10 covers a restricted trial; it does not establish that 288 runs, their review and the corrections will fit. Estimate preparation, collection, answer review and content production separately.

| Review | Decision to prepare | Evidence to retain |
|---|---|---|
| Day 30 | Is the baseline sufficient to begin the trial? | Panel, initial answers, verified access and urgent corrections already made |
| Day 60 | Can the change and comparison be assessed? | Page versions, answers to the same questions and actual time spent |
| Day 90 | Continue, narrow the scope, extend observation or stop? | Qualified enquiries, known outcomes, costs, open questions and the next time limit |

An open sales opportunity retains that status at the review. A sale during the trial is not automatically an outcome caused by the intervention. If a technical correction consumes the budget, adjust the editorial scope and record the decision.

### I. A weekly review in five decisions

At each review, consider in order: a factual error to correct; a confirmed access obstacle; a buying question to cover better; a relevant external contribution; and a decision about continuing based on the available results. The owner may decide to add nothing when evidence is missing. Keeping the date, cost, and reason for the decision helps avoid rewriting the project’s history after a good or bad result.