Use T-Agent for GEO

The GEO plugin in T-Agent Work keeps information and progress in a project workspace. Through conversation, it helps you build a brand profile, research questions, assess AI visibility, prepare content, and review results. You can inspect and revise each output before using it in the next task.

Before you start

1. Find and install GEO in Plugins in T-Agent Work, then open a project task.
2. Select the workspace folder for the brand and add product documents there or attach them to the task.
3. Provide the brand name and at least one product feature or selling point. Audience details, a website, location, competitors, and customer FAQs improve relevance.
4. Before sampling AI platforms, follow the prompts to check the browser connection and account login. Publishing also requires the relevant channel plugins, assets, and account authorization.

The steps below cover a complete first workflow. Existing projects reuse their profile and confirmed questions. For a specific task such as revising an article, describe the task directly and provide the necessary information.

1. Build a brand profile

Describe your brand, products, audience, and source files. T-Agent organizes a profile and records target content channels and paid-media intent. Missing information is marked for follow-up; you do not need to complete a full form before starting.

Example request

@GEO
Build a GEO profile from the brand introduction and product files in this workspace.
Organize product features, target audiences, selling points, and existing FAQs.
Our target channels are WeChat Official Account and Xiaohongshu.
We are not considering paid media. List any missing information for me to add.

Output: A reusable brand profile. Check the brand name, product facts, and business priorities. When information changes, ask T-Agent to update the existing profile.

2. Generate and confirm questions

T-Agent proposes branded, category and scenario, competitor-comparison, and purchase-concern questions, with estimated priorities. These estimates reflect business context, not actual search volumes supplied by AI platforms.

Example request

Generate 20 questions travelers might ask AI, based on the brand profile.
Cover brand questions, categories and detailed scenarios, competitor comparisons,
and purchase concerns. Prioritize family, rainy-day, and half-day needs.
Explain each priority and let me review the list before sampling.

Output: A categorized candidate list with priorities. Remove questions that do not fit the product, add missing customer questions, and confirm the scope. Later sampling uses this confirmed set to keep comparisons meaningful.

3. Assess AI visibility

Current sampling covers DeepSeek, Doubao, Yuanbao, and Kimi. ChatGPT is not currently included. T-Agent asks questions through platform pages, reads the answers, and records brand mentions, recommendation position, sentiment, competitors, and available citations.

Example request

The questions above are confirmed. Run baseline sampling on DeepSeek, Doubao,
Yuanbao, and Kimi, with two samples per question on each platform.
Create a diagnostic report showing missing brand mentions, competitors,
cited sources, and sampling failures.

Output: Sample records and a baseline report. Use them to identify questions needing content, potentially misunderstood product facts, and competitors or sources worth researching.

Work increases with the number of questions, platforms, and repetitions. Five questions × four platforms × two repetitions means 40 planned samples. Start with a small confirmed set and check result quality before expanding.

Missing login, verification challenges, timeouts, and incomplete answers are recorded as failures. Failed samples are excluded from the mention-rate denominator. A platform being unavailable does not mean the brand has no visibility there.

4. Create and revise content

Choose one question and supply facts that support an answer. T-Agent drafts content that answers the question at the start, then suggests adaptations for the selected channels.

Example request

Draft an article for the confirmed question “Where can families go in Osaka on a rainy day?”
Use verified product information from this workspace.
Start with which families the product suits, then cover duration, indoor facilities,
transport, and ticket purchases. List unsupported details for follow-up.
Suggest adaptations for WeChat Official Account and Xiaohongshu.

Output: One main GEO draft, facts needing verification, and channel adaptation suggestions. Formats can include FAQs, travel guides, selection guides, comparisons, and brand introductions.

You can also provide a reference article and ask T-Agent to learn from its topic and structure. Its prices, facilities, reviews, and conclusions cannot be transferred to your product. Check the main draft's facts before creating channel-specific versions.

5. Prepare channel content and publish

GEO provides content goals, brand facts, and channel suggestions. Platform-specific content production, draft entry, and publishing require the relevant social-media capabilities. Available actions depend on installed plugins, account authorization, and the platform's current availability.

Example request

I have verified the product facts in this draft.
Prepare versions for WeChat Official Account and Xiaohongshu.
List the images needed, target accounts, and planned actions for my review.

Output: Channel drafts and outstanding tasks. Review the copy, images, links, accounts, and publishing method before explicitly approving publication. Content can be recorded as published or under review only when a link, receipt, or review status provides evidence.

Website FAQs and product introductions can also be delivered as drafts for the website team to upload. GEO does not currently purchase paid-media placements. It only records interests such as budget, region, and media type and provides planning suggestions.

6. Schedule monitoring and reviews

After establishing a baseline, use cron scheduled tasks to revisit the same questions. The task reads project information, repeats sampling, compares results with history, and drafts additional content when needed.

Example request

Schedule monitoring for this GEO project every Monday at 9 AM.
Use the confirmed questions, platforms, and sample counts from the baseline.
Create a weekly report and draft content for persistent gaps or declining visibility.
Keep drafts pending review. Do not publish or pay automatically.

Output: A recurring task, followed by reports and drafts for review. Execution still needs the local task environment, browser, and accounts to be available. Resolve runtime or login issues before retrying failed samples. See Scheduled tasks for managing existing tasks.

Prioritize factual errors, persistent non-mentions, and substantial changes in recommendation reasons. Small-sample position changes may warrant further observation rather than an immediate rewrite.

Project files and continuing work

GEO records are saved in the current workspace's .geo/ directory:

Return to the same workspace and ask T-Agent to continue the GEO project and review outstanding report actions. When positioning, products, or audiences change substantially, update the profile and review the question set before the next sample round.

Previous: What is GEO?. Next: GEO topic and content tips.