# Portfolio Project: Rewrite a Page for GEO and Measure It

> A step-by-step portfolio project: rewrite one page for AI answers, test before and after, and present the result honestly.

Source: https://mentionx.ai/learn/careers/portfolio-project-rewrite-page-geo-measure  
Author: Dipuck Jones, founder of MentionX  
Updated: 30 October 2026  
Level: Practitioner

**A strong GEO portfolio project takes one real page, records how AI engines answer related questions, rewrites the page so it answers those questions clearly, then re-tests after the engines have had time to see it. The rewrite matters less than the method: a fixed prompt set, several runs, dated results and an honest write-up of what changed and what didn't.** Use a page you own or have permission to change.

*Checked 3 Nov 2026 (CET). "Our view" marks opinion. Examples are fictional.*

> **Key takeaways**
>
> - Pick a page you control and can keep live for several weeks.
> - Record a dated baseline with fixed prompts, several runs and two or three engines.
> - Rewrite for a direct answer, clear headings and visible facts, and log every change.
> - Re-test with the same method and report counts, small samples and limits honestly.

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

## Why is this a good portfolio project?

Hiring managers for GEO roles want evidence that you can do the work, not just talk about it. A before-and-after page project shows the whole loop: research, writing, technical basics, measurement and honest reporting. It's small enough to finish in a few weeks and concrete enough to discuss in an interview.

It also shows judgment. Anyone can rewrite a page. Fewer people can explain why they changed what they changed, set up a fair test, and say clearly that results were mixed when they were. *Our view:* an honest "partly worked" write-up impresses more than a suspiciously perfect one.

## How do you choose the page?

Pick a page you control: your own site, a side project, a friend's business or a volunteer organisation that agrees in writing. You need to be able to publish changes and keep them live for several weeks. Don't edit pages you don't own, and don't create fake businesses to game results.

Good candidates answer a clear question people ask an AI assistant: a how-to guide, a service page, a product comparison or a local business page. Avoid pages with almost no search interest, because engines may never surface them, and avoid pages that are already the obvious answer, because you'll have no room to improve.

## How do you set the baseline?

Write eight to twelve prompts a real person might ask where this page should be a good answer. Mix direct questions, comparisons and problem questions. Run each prompt three times in two or three engines, signed out where possible, and record the date.

For each run, note whether the page or brand is mentioned, whether it's cited, what's said and which other sources appear. Save screenshots or copies of answers. Also record the page's search performance from your search console if you have it, so you can see whether traditional search changed too.

## What should the rewrite change?

Start the page with a direct answer to the main question in two or three sentences. Use descriptive headings phrased the way people ask. Put key facts, such as prices, steps, requirements or locations, in plain text rather than inside images. Add clear authorship and a last-updated date. Remove filler that delays the answer.

Check the basics. The page should be indexable, load without relying entirely on JavaScript for its main text, and have an accurate title and description. If it suits the page, add structured data that matches visible content. Don't add claims you can't support or hidden text aimed at machines. Google says its usual guidance on helpful, people-first content applies to AI features too.

Keep a change log: what you changed, when and why. Note anything else that changed on the site during the project, because it could affect results.

## How long should you wait before re-testing?

Wait until you've confirmed the page has been recrawled, through your search console's URL inspection or server logs, and then give it a few more weeks. Engines that search the live web may pick up changes quickly; answers driven by model training may not change during your project at all. Say this in your write-up.

Re-run exactly the same prompts, the same number of times, in the same engines, under the same conditions. Changing prompts mid-project breaks the comparison.

## How do you present the result?

Use a short case study: the question, the page, your baseline, what you changed, the re-test and what you learned. Show counts, such as "cited in 2 of 9 runs before, 5 of 9 after, in one engine," and keep the small sample size visible. Include one or two before-and-after answer excerpts.

Then explain what you can and can't conclude. With small samples and other changes happening on the web, you can show a direction, not prove a cause. Say what you'd test next. That paragraph is often what interviewers ask about.

## What should you avoid?

Don't exaggerate. Don't hide runs that went badly. Don't publish screenshots that include other people's private data. And don't present a project on a client's site without written permission. If you can't share details, describe the method and anonymise the results.

## From MentionX data: what engines tend to cite

**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.

MentionX's anonymised aggregates show which kinds of page engines cited across 9,750 citations:

| 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% |

*Our view:* blog and article pages make up the largest share of citations in this sample, so a well-structured guide is a sensible choice for your project page.

**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: Ledgerline's invoice-approval guide

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

A candidate volunteers to help Ledgerline, a fictional accounts-payable tool, with its guide "How to set up invoice approvals." Her baseline uses 10 prompts, three runs each, in two engines. The guide is cited in 1 of 60 runs.

She moves a three-sentence answer to the top, turns the steps into a numbered list, adds a table of approval rules and a dated author line, and fixes a title that didn't match the content. After confirming a recrawl and waiting four more weeks, the same test shows citations in 6 of 60 runs, almost all in one engine. Her write-up says the change is encouraging, the sample is small and one engine didn't move at all. That honesty is what the interviewer asks about first in this illustrative example.

> **Pro move:** Put the exact prompt list in an appendix. It lets reviewers judge your test design and shows you expect scrutiny.

## What can you do this week?

Choose your page, write your prompts and run the baseline today, before you change anything. A baseline you skipped can't be recreated later. Then plan the rewrite and book a calendar reminder for the re-test. For other project ideas, see [building a GEO portfolio](https://mentionx.ai/learn/careers/building-geo-portfolio-5-projects-get).

## How do you make the test fair?

Control what you can. Run baseline and re-test prompts at a similar time of day, in the same browser state, with the same account settings. Avoid changing other pages on the site during the project, or at least log any changes. If an engine updates its model or interface during your test window, note the date, because it may explain shifts that have nothing to do with your rewrite.

Consider a simple comparison page too. If you have a second, similar page you don't change, test it alongside your project page. If both improve, something other than your rewrite may be responsible. If only the rewritten page improves, your case is stronger. You still can't prove cause with a sample this small, but a comparison page makes your reasoning more convincing.

## What should your case study include?

Use a consistent structure so reviewers can scan it:

1. The question the page answers and why it matters to the audience.
2. The prompt set, engines, run count and test dates.
3. Baseline results, with counts and two example answers.
4. The change log, with dates.
5. Re-test results using the same format.
6. What you conclude, what you don't and what you'd test next.

Keep it to two or three pages, with your raw sheet linked. Write in plain language and avoid jargon a hiring manager outside SEO might not know. If you can, include a short note on what you learned that surprised you. Interviewers often remember that part.

## How do you talk about it in an interview?

Prepare a two-minute version: the problem, the method, the result and the lesson. Then be ready to go deeper on any step. If asked whether the rewrite "worked," answer with the counts and the caveats. Showing that you know the difference between a promising signal and a proven result is exactly what employers hiring for GEO roles look for, because so much of the field is still unsettled.

## Which write-up gaps cost candidates the most?

The most common is claiming too much. Saying a rewrite "increased AI visibility by 500%" because citations went from one to six in a small sample will make an experienced reviewer distrust everything else. Report the counts and let the reader judge.

The second is leaving out the baseline detail. If you don't say how you tested before, nobody can tell whether the after numbers mean anything. Third is changing the prompts between runs, which turns your comparison into two separate tests. Fourth is hiding failures. If one engine didn't change at all, say so and offer a reason you'd want to test, such as that engine answering mostly from training data rather than live search.

Finally, keep the project going if you can. Re-running the same test three months later and adding a short update shows patience and curiosity, both qualities hiring managers look for in a field where results take time.

## How do you keep the project ethical?

Be open with the page owner about what you're testing and what you'll publish. Don't seed fake reviews, forum posts or links to influence results, because that misleads real people and can break platform rules. Don't run thousands of automated prompts against an assistant if its terms forbid it. Keep personal data out of screenshots, and credit anyone who helped. A clean, transparent project is something you can discuss proudly in any interview; a clever shortcut is something you'll have to explain away. If you're unsure whether a tactic is acceptable, ask whether you'd be comfortable describing it in detail to the hiring manager.

## Common questions

**Can I use a client's page?** Yes, with written permission. Otherwise anonymise it or use a page you own.

**How long does the project take?** Usually several weeks, because you need to wait for recrawling before re-testing.

**What if nothing changes?** Report it honestly and explain what you'd try next. That's still a useful case study.

**How many engines should I test?** Two or three is enough for a portfolio project.

## Sources

- Google Search Central, [AI features and your website](https://developers.google.com/search/docs/appearance/ai-features)
- Google Search Central, [Creating helpful, reliable, people-first content](https://developers.google.com/search/docs/fundamentals/creating-helpful-content)
- Google Search Console Help, [Performance report](https://support.google.com/webmasters/answer/7576553)
All accessed 3 Nov 2026.

## Frequently asked questions

### Can I use a client's page?

Yes, with written permission. Otherwise anonymise it or use a page you own.

### How long does the project take?

Usually several weeks, because you need to wait for recrawling before re-testing.

### What if nothing changes?

Report it honestly and explain what you'd try next. That's still a useful case study.

---

Written and reviewed by Dipuck Jones for MentionX Learn. The HTML version of this page: https://mentionx.ai/learn/careers/portfolio-project-rewrite-page-geo-measure