# Capstone: GEO Practitioner

> GEO Practitioner capstone: run a 60 day GEO sprint with a baseline and rescan, citation gap analysis and a monthly report dashboard.

Source: https://mentionx.ai/learn/paths/geo-practitioner/capstone  
Author: Dipuck Jones, founder of MentionX  
Updated: 2 November 2026  
Level: Practitioner

**The GEO Practitioner capstone is a 60 day GEO sprint: you set a baseline, run a focused set of technical, content and citation changes, then rescan and report what changed.** You'll submit a sprint plan, baseline and rescan data, a citation gap analysis and a monthly report dashboard. The rubric rewards sound method and honest analysis; a positive result isn't required to pass.

> **Key takeaways**
>
> - The capstone is a 60 day sprint: baseline, technical check, citation gap analysis, focused changes, rescan and a report.
> - Keep the method identical between scans, use repeat runs and log every change with its date.
> - Report change with ranges and name confounders; the rubric rewards honest analysis, not a positive result.
> - You pass with 70 out of 100; the analysis and honesty criterion alone is worth 20 points.

*Checked on 31 October 2026. "Our view" marks MentionX opinion. Ledgerline is a fictional company used for illustration.*

## What's the brief?

1. **Pick a scope.** One brand, one priority topic and 30 to 50 prompts. A narrow scope gives clearer results.
2. **Run a baseline.** Scan the prompts across at least 3 engines, ideally with repeat runs, in a MentionX free account. Record mention rate, citation rate, recommendation rate and share of voice versus competitors.
3. **Do a technical check.** Confirm AI crawler access in robots.txt, firewall and CDN settings, rendering and server logs, using [Technical GEO](https://mentionx.ai/learn/technical/list-ai-crawlers-user-agents).
4. **Run a citation gap analysis.** List the URLs engines cite for your prompts, then classify which ones mention you, which mention only competitors and which you could realistically influence.
5. **Write a 60 day sprint plan.** Include 5 to 10 prioritised actions with owners, dates and the metric each should move.
6. **Execute and log.** Record every change with its date so you can match it against results.
7. **Rescan at day 60** with the same prompts, engines and method.
8. **Build a monthly report dashboard** showing baseline versus rescan, with confidence intervals or ranges where relevant.

## What must you submit?

| Item | Contents |
|---|---|
| Sprint plan | Scope, prompt set, actions, owners, dates, target metrics |
| Baseline and rescan data | Exports with dates, engines and run counts |
| Citation gap analysis | Cited URLs classified by influence and status |
| Change log | Every action with date and link |
| Report dashboard | Baseline vs rescan, ranges, interpretation, next steps |

## How is it scored?

| Criterion | Points | What earns full marks |
|---|---|---|
| Measurement design | 20 | Consistent method across both scans, repeat runs, competitors included |
| Technical audit | 15 | Crawler access verified with evidence (robots.txt, logs or tests) |
| Citation gap analysis | 20 | Sources classified, realistic targets chosen |
| Execution | 15 | Actions completed and logged with dates |
| Analysis and honesty | 20 | Change shown with ranges; noise and confounders discussed; no causal overclaiming |
| Reporting | 10 | Clear dashboard a stakeholder can read in two minutes |
| **Total** | **100** | **Pass: 70** |

## What does a strong report look like?

*Fictional example only; numbers are invented to show the format.*

> **Scope:** 40 prompts on "invoicing software for freelancers", 3 engines, 3 runs per scan.
>
> **Baseline → day 60:** mention rate 18% → 27%. With 360 answers per scan, the 95% intervals roughly overlap at the edges, so we report this as a likely but not certain improvement.
>
> **What changed:** Perplexity began citing our refreshed comparison page in 9 answers. ChatGPT showed no clear change.
>
> **Confounders:** a competitor's pricing page went offline in week 5, which may have helped.
>
> **Next:** extend the comparison format to two more topics; continue pitching two cited review sites.

## How should you plan the 60 days?

| Days | Focus |
|---|---|
| 1 to 7 | Scope, prompts, baseline, technical check |
| 8 to 14 | Citation gap analysis and sprint plan |
| 15 to 50 | Execute and log |
| 51 to 60 | Rescan, analysis and dashboard |

## From MentionX data: why the source mix matters for your gap analysis

**Method.** Read only aggregate queries over MentionX scans from 22 August to 29 October 2026 (UTC): 2,217 engine runs (2,154 completed) across 18 anonymised projects and six engines.

Your citation gap analysis will classify the URLs engines cite. In our sample of 9,750 citations, the types of source cited looked like this:

| Cited URL type | Share of 9,750 citations |
| --- | --- |
| Blog or article | 19.6% |
| Homepage | 10.0% |
| Review page | 2.7% |
| Docs | 2.6% |
| Pricing | 1.9% |
| Wikipedia | 1.6% |

Engines also differ in how often they cite at all:

| Engine | Brand mentioned | Answers with no citations |
| --- | --- | --- |
| Gemini | 47.3% | 50.7% |
| OpenAI | 42.9% | 30.2% |
| Perplexity | 31.5% | 4.1% |
| AI Mode | 31.3% | 12.6% |
| Copilot | 26.0% | 47.0% |
| DeepSeek | 24.0% | 99.0% |

Two things follow for the capstone. First, expect your list of cited URLs to come mostly from the engines that cite heavily; engines with high no citation rates will add little to the gap analysis. Second, the mix of source types tells you where effort is realistic. You can change your own pages quickly; earning a place on a third party list or review site takes longer and may not show within 60 days.

*Our view:* in a 60 day sprint, weight your plan towards sources you control or can influence quickly, and treat slower third party work as groundwork for the next sprint.

**Limits.** A convenience sample of mostly B2B software projects, run through engine APIs, which can behave differently from the consumer apps. Perplexity, Copilot and DeepSeek figures come from a single day (26 October 2026), so treat them as a snapshot, not a trend.

## Worked example: a Ledgerline sprint plan

*Ledgerline is a fictional accounts payable tool. Figures are illustrative.*

A learner practises on Ledgerline, a fictional accounting software company, competing with Competitor A, B and C. Scope: 40 prompts about invoicing for small agencies, three engines, three runs per scan. The baseline shows Competitor A recommended in most comparison answers, often citing a review site and a listicle where Ledgerline is missing or outdated.

The sprint plan has six actions: allow the search crawlers blocked by an old firewall rule, refresh the agency invoicing page with an answer first summary and comparison table, add a pricing FAQ, update the review site profile, pitch the listicle author with current facts and publish one original guide. Each has an owner, a date and the metric it should move.

> **Pro move:** Give each action a single target metric. If an action has no metric, ask whether it belongs in this sprint.

## How do you pick a scope that works?

Narrow beats broad. One brand, one priority topic and 30 to 50 prompts give you enough data to see change and few enough moving parts to explain it. Pick a topic that matters commercially and where the brand is beaten but not absent; that's where a 60 day sprint is most likely to show movement.

Write prompts the way buyers ask, and include comparisons and problem led questions as well as category ones. Freeze the list before the baseline. Every later change to the prompt set weakens your comparison.

## How do you run a reliable baseline?

Run every prompt on at least three engines, with repeat runs, ideally three per prompt per engine. Record the date, engine, whether you were signed in and the run number. Track four metrics: mention rate, citation rate, recommendation rate and share of voice versus competitors. Capture cited URLs in full; they feed the gap analysis.

Repeat runs let you estimate noise. If the same prompt gives different answers across runs, that variation is your baseline uncertainty, and any later change needs to be larger than it before you can call it real.

## What should the technical check prove?

The technical audit is worth 15 points and needs evidence, not assertions. Show the relevant lines of robots.txt, confirm that firewall and CDN rules don't block AI crawlers you want to allow, check that key pages render without heavy client side scripts and, if you can, show crawler visits in server logs. If you find a block, fixing it is often the highest value action in the sprint, so log the fix date carefully.

## How do you build the citation gap analysis?

List every URL cited for your prompts. Classify each one: does it mention you, only competitors or neither? Then rate how realistic it is to influence: your own pages (high), profiles and directories you control (high), review sites (medium), lists and articles by others (medium to low), forums (low and slow, and only with honest participation). Choose targets where influence is realistic and the URL appears often.

## How do you execute and log changes?

Keep a change log with the date, URL and a one line description for every action. Include things outside your control that you notice, such as a competitor's site change or a news story. The log is what lets you discuss confounders honestly at the end, and graders check that it matches your plan.

## Where do sprint submissions go wrong?

- **Changing the method between scans.** Different prompts or engines make comparisons meaningless.
- **Single runs.** AI answers vary; repeat runs show how much.
- **Too many changes at once.** Log everything and be modest about attribution.
- **Causal claims.** Say "coincided with", not "caused", unless you have a control.
- **Hiding flat results.** A flat result with good analysis can still score well.

## How do you analyse the rescan?

Run the same prompts on the same engines with the same number of runs and the same signed in state. Compare each metric with its baseline and show ranges, such as 95% intervals, where you can. Look at which prompts and engines moved and check them against your change log. If a refreshed page started appearing in citations, that's strong evidence; a general rise across unrelated prompts is weaker.

## What makes the dashboard readable?

A stakeholder should understand it in two minutes. Put the headline metrics, baseline versus rescan with ranges, at the top. Below, show what changed by engine, the actions completed and the confounders. End with next steps. Avoid dense tables at the top; save detail for an appendix.

## How do you handle confounders honestly?

Over 60 days, plenty happens that you didn't do. Engines update their models and features, competitors publish or remove pages, news stories shift what people ask and seasonal demand changes the questions themselves. None of that is a reason to give up on measurement, but it is a reason to be careful about what you claim.

List the confounders you noticed in your change log and say how each might have affected the results. Where you can, look for evidence that separates your actions from outside events. A page you changed that starts being cited is specific evidence; a rise in mentions across prompts you didn't work on suggests something broader. If a competitor's page disappeared, check whether the answers that changed are the ones it used to win.

*Our view:* graders give more credit for a careful "we can't be sure" than for a confident claim the data doesn't support. Real stakeholders do too, once they've been burned by an overclaim.

## What should your next steps section say?

End with what you'd do in the next 60 days. Keep the actions that showed promise, drop or change the ones that didn't and add the slower third party work you started. Suggest any changes to the method, such as more runs or an extra engine, that would make the next comparison stronger. A good next steps section shows that the sprint was one turn of a repeatable cycle, not a one off project, and that's the habit this capstone is meant to build. Write it so that someone else could pick up the work tomorrow and continue it without asking you a single question about what you meant. That's the standard to aim for.

## Common questions

**What if my metrics don't improve?** You can still pass. Explain the result honestly, discuss noise and confounders and propose what you'd change.

**Do I need statistics software?** No. A spreadsheet can calculate simple intervals for rates.

**Can I include more than one topic?** Keep to one. Extra topics make the results harder to interpret.

**How many runs per prompt are enough?** Three per engine per scan is a sensible minimum for a sprint of this size.

## Where can you get help?

Use [Measuring AI Visibility and ROI](https://mentionx.ai/learn/measurement/calculating-roi-of-geo) for metrics and intervals, [Earning AI Citations](https://mentionx.ai/learn/citations/earned-media-owned-content-geo) for the gap analysis and the [citation gap analysis template](https://mentionx.ai/learn/templates/citation-gap-analysis-template) and [monthly reporting dashboard](https://mentionx.ai/learn/measurement/monthly-ai-visibility-report-template) for structure.

A passing sprint report belongs in your portfolio. Next, take the [Lead AI Search Strategy path](https://mentionx.ai/learn/paths/lead-ai-search-strategy).

## Frequently asked questions

### Why does it take 60 days?

AI answers change slowly and vary between runs. Sixty days gives changes time to be crawled and gives you enough repeat data to see a signal.

### Do I need a paid tool?

No. A MentionX free account or a disciplined manual process is enough.

### What if visibility doesn't improve?

You can still pass. The rubric rewards sound method and honest analysis, not a particular result.

---

Written and reviewed by Dipuck Jones for MentionX Academy. The HTML version of this page: https://mentionx.ai/learn/paths/geo-practitioner/capstone
