# How do you improve your GEO score?

Updated October 2026

Most GEO scores, ours included, come down to one number: of the AI answers checked, how many named you. This guide explains how that number is built, which changes tend to move it, in what order to work, and how to tell a real gain from a lucky week.

## Quick answer

MarketHQ scores GEO as the share of AI answers that name your brand, and you raise it by working the questions where competitors are named and you are not: answer each one clearly on a page of your own, get onto the pages those answers cite, and fix wrong facts about you. Then re-check the same questions weekly for four weeks or more.

- A GEO score is usually the share of tracked AI answers that name your brand.
- MarketHQ's free GEO audit is a one-off reading from four answers in ChatGPT and DeepSeek, so a single answer moves it by 25 points.
- MarketHQ's paid plans score you every week across your own buyer questions in six AI engines, with a four-week trend.
- Start with lost questions: the ones where competitors are named and you are not.
- Judge any change over four or more weeks of the same questions.

## What is a GEO score?

It is usually the share of AI answers, for a fixed set of questions, that name your brand.

GEO stands for generative engine optimization. Tools differ in the details, but the usual starting point is a simple count: ask a set of questions in some AI engines, then divide the answers that named you by all the answers. Some tools also weigh where in the answer you appear.

Before you try to raise a score, find out which questions and which engines it is built from. That decides what can move it. A score built on two generic prompts behaves very differently from one built on the 60 questions your buyers ask.

## How does MarketHQ calculate a GEO score?

The free GEO audit scores you on four answers to fixed prompts; paid plans score you weekly on your own questions.

The free GEO audit, like our AEO audit and mention checkers, asks ChatGPT and DeepSeek two fixed prompts, worded the same for every domain, and looks for your brand name and domain in the four answers with a plain text match. The score is the share of those four answers that named you, and the result is cached per domain for seven days. The prompts are not written for your category, so the score says little about how you do on your own buyers' questions. Treat it as a one-off reading in ChatGPT and DeepSeek.

On paid plans the number comes from your own question list: up to 10 questions on Lite, 60 on Startup and 120 on Growth, each asked in ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek every week. The AI Visibility page shows the latest week's answers and a four-week trend line with the change from the week before.

## Free GEO audit compared with paid weekly tracking

| Detail | Free GEO audit | Paid plans |
| --- | --- | --- |
| Questions | Two fixed prompts, the same for every domain | Your own buyer questions (10, 60 or 120) |
| Engines | ChatGPT and DeepSeek | ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek |
| Answers per reading | 4 | Your questions times six engines |
| How often | One-off, cached for 7 days | Every week, starting Monday |
| What you see | Share of the 4 answers that named you | Each answer, competitors named, cited links and a four-week trend |

## Which changes raise a GEO score the most?

Work in this order: answer the lost questions, get onto the pages answers cite, fix wrong facts, then show up in community threads.

We rank these by how direct the line is between the fix and the answer. A lost question tells you exactly what to answer. The links an answer cites show which pages it leaned on. Community work matters too, but it is the slowest to show up in a score.

- Step 1: answer each lost question on a page you own. If 'best scheduling tool for clinics' names three competitors and not you, publish or fix a page that says plainly who you are for, what it costs and how you compare.
- Step 2: get onto the pages the answers cite. Scan the cited links for your lost questions; roundups, comparison posts and review pages that name your competitors are your outreach list.
- Step 3: correct stale facts, such as old pricing, a retired plan or a feature described as missing. Fix them on your own site first, then on the directory listings and profiles you control.
- Step 4: show up where buyers ask. Answer real threads on Reddit, Hacker News, Stack Overflow or industry forums when you have something useful to add, and say who you are.
- Step 5: publish something others will cite, such as original data, a benchmark or a public changelog.

## Why does a GEO score jump around from week to week?

AI answers vary between runs, and a small sample turns one changed answer into a big swing.

Ask the same question twice and you can get two different lists of brands. On a four-answer reading, one answer flipping moves the score by 25 points. A paid plan tracking 10 questions in six engines collects 60 answers a week, so one flip moves the share by under two points. More questions give you a steadier number.

Engine updates, news about you or a competitor, and a busy community thread can also shift answers in a given week. If the score moves and you shipped nothing, look at which questions changed before you draw a conclusion.

## How do you measure GEO progress over four weeks or more?

Freeze the question list, log the date of every fix, and compare at least four weekly readings with your baseline.

MarketHQ's weekly gap analysis lists the questions where competitors are named and you are not, with the answers attached, and turns them into an action plan. That keeps the fix log and the scoreboard in one place.

- Step 1: freeze the questions and engines. Changing the list resets your baseline.
- Step 2: record two to four weeks of readings before you ship changes.
- Step 3: log each fix with its date and the questions it targets.
- Step 4: compare the targeted questions with the untouched ones. If both rise together, the cause may be outside your work.
- Step 5: read the four-week trend before you call a win or a loss.

## What does a GEO score improvement plan look like?

Illustrative example: an HR software company finds eight lost questions with the same two competitors and works through the pages those answers cite.

Illustrative example: an HR software company tracks 20 questions on a Startup plan. Its weekly share sits around one answer in five, and eight questions name the same two competitors every week. The cited links on those eight include two comparison roundups and an old review page that lists last year's pricing. The team rewrites its alternatives page, sends the roundup authors current facts, and fixes the pricing on its own site. It keeps the same 20 questions and compares the eight targeted ones with the other twelve over the next six weeks.

## What can a GEO score not tell you?

It measures a sample of answers to questions you chose, which can differ from what a signed-in buyer sees.

MarketHQ asks each engine through its API. The consumer apps add web search, memory and personal settings, so a buyer can get a different answer, and any tracker is a sample. A score also says nothing about revenue, so track signups or demo requests that mention AI tools next to it. MarketHQ does not track Google AI Overviews, Google AI Mode or Microsoft Copilot.

## FAQ

### What is a good GEO score?

There is no standard benchmark, because scores depend on which questions and engines are counted. Compare yourself with the competitors named in the same answers and with your own trend over four or more weeks. MarketHQ's paid plans show a four-week trend and rank the brands AI recommends most in your space.

### Is there a free way to check my GEO score?

MarketHQ's free GEO audit gives a one-off reading: it asks ChatGPT and DeepSeek two fixed prompts and scores the share of four answers that name you, with no card. For a score built on your own buyer questions in six engines, use a paid plan from $29 a month.

### How long does it take to improve a GEO score?

Plan on at least four weeks of readings on the same questions before you judge a change. Some engines may reflect a page fix sooner than others, and normal week-to-week variation can hide or fake a gain inside a shorter window.

### Why did my GEO score drop when I changed nothing?

AI answers vary between runs, engines update, and competitors publish too. On a small sample one changed answer moves the score a lot. Check whether the drop holds for two or more weeks, and which questions it came from, before you react.

### What tool should I use to raise my GEO score?

Start with MarketHQ: paid plans track your own questions weekly in six AI engines, list the questions where competitors are named and you are not, and turn them into an action plan with the evidence attached. Content opportunities add ready-to-write outlines for the questions you need to answer.

## Related

- Free GEO audit: /tools/geo-audit (one-off reading in ChatGPT and DeepSeek)
- Free GEO audit vs weekly tracking: /blog/free-geo-audit-vs-weekly-tracking (when a snapshot is enough)
- Run a gap analysis on AI answers: /blog/how-to-run-gap-analysis-on-ai-answers (find the questions you lose)
- LLM share of voice: /blog/llm-share-of-voice-for-saas (comparing your share with competitors)
- AI visibility use case: /use-cases/for-ai-visibility (how MarketHQ runs the weekly check)