Geova AI Chat
Purpose: Use the detailed Geova AI controls within Mapplex.
Before you start: The intended project, a signed-in account, enabled workspace AI access, an internet connection and available credits.
Expected result: You can inspect the scope, reasoning, evidence, and proposed next action.
Geova AI helps you query project data in plain language. It can draft SQL filters, summarize records, prepare exports, and preview spatial results. Review generated queries and results before using them for decisions.
1. Opening Geova AI
On a phone, open the menu and choose Geova AI Partner. On a tablet, choose Geova AI in the side navigation rail. Confirm the active project before asking a question. Follow the global AI review for worked project questions. The chat interface opens with:
- Header — Geova AI, private-chat control, clear history and close.
- Project context — Expand this row to check the project totals, layers and active layer.
- Message Area — A scrollable chat history
- Suggestions & mentions (+) — Expand beside the input to see query suggestions and layer/column mention controls
- Input Area — Text input for typing queries, with
@and#mention support
Phone and tablet controls
Expand Project context to check your data and Suggestions & mentions (+) to find prompts and layer names. On a wider screen, collapse the navigation rail to leave more room for the conversation. The practice questions use fictional project data.
Choose the Geova AI tool for the task
| Task | Where to start | Useful result |
|---|---|---|
| Understand existing project data | Geova AI Chat | A scoped answer, evidence, and result controls |
| Design a new collection project | Open Geova Project Planner, where enabled | A proposed design, secure review, and explicit creation step |
| Prepare field choices | The Lexicon panel in Forms | A reviewable list to save and link |
| Draft conditional form behavior | Simple Smart Logic's AI prompt | A rule to inspect and test in the form |
| Review a suggested answer for one field | A supported focused form co-pilot card | A one-field proposal with separate Apply/Undo checks |
Planner requires enabled Teams/Enterprise access and suitable role permissions; hosted-project capacity is a separate allowance. Geova AI chat can review permitted work and propose creating or assigning asset work orders. Nothing changes until you choose Confirm on the proposal, and your role must allow the action. Creating a project still uses Planner's separate review and creation steps. See Review a Work proposal. See Project Planner and Forms & Smart Logic for those workflows.
2. Asking Your First Question
Follow Questions to ask your global project for four worked question-and-answer examples.
Typing a Query
- Type a question in the input area:
"How many manholes are in Zone A?"
- Tap the Send button (or press Enter).
- A loading state appears while the request is processed.
- Geova AI responds with a formatted answer, often including record counts, tables, or map previews.
Using Query Suggestions
Expand Suggestions & mentions (+) beside the input to find query chips based on your project's schema. For example, if you have a layer called "Fire Hydrants" with fields like "Condition" and "Flow Rate":
- "Count all Fire Hydrants"
- "What is the average Flow Rate?"
- "Group Fire Hydrants by Condition"
Tap a chip in the horizontal Suggestion Bar to fill the input, then review or edit it and choose Send. The larger Suggested Queries buttons on the welcome screen submit immediately. With a keyboard, Enter sends and Shift+Enter adds a new line.
Short Questions and Supported AI Languages
Geova AI question understanding currently supports English, Arabic, and mixed English/Arabic GIS terms. Other interface languages do not yet expand the AI question-understanding contract.
For a uniquely named project layer, short questions can be enough. The names below are illustrative placeholders, not built-in Mapplex layers or fields; choose names from your active project:
roadscount roadsعدد مواقع الفحصقيم الحالة في مواقع الفحصshow damaged roadscount roads by districtwhat statuses exist in roadshighest length in roadsاعرض roads حيث status تالفاحسب عدد roads حسب district
Use the actual layer, field and stored choice names. Mentions and configured choice lists help Geova interpret the question. If a word is unclear, Geova may ask you to choose a value or provide more detail.
Check that the answer used the intended field and condition. For example, filtering a City field is different from finding points inside a city boundary. If the answer uses the wrong operation or unexpected records, clarify the question before saving or exporting the result.
Welcome Screen
If no messages exist, the Empty State shows:
- Welcome message: "Ask Me Anything"
- Informational badges: Limited samples may be sent and 1 credit per useful answer. These labels do not mean that all AI processing happens offline; read Privacy & Data Safety.
- Local project-guide helpers: Explore Data, Layers, Fields, and Spatial. These describe available data and possible questions; they are not completed analytical results.
- Suggested Queries based on your project data
Use Explore Data when you are not sure what to ask. Geova AI lists the layers, record counts, useful fields, and example questions it can answer from the active project.
3. Mentions: @Layers and #Columns
Mentioning a Layer
Type @ in the input field to open the Layer Mentions Dropdown. This shows all layers in the active project. Select a layer to reference it in your query. The following names are examples only:
"Show all records in @Fire_Hydrants where Condition is Poor"
Mentioning a Column
Type # to open the Column Mentions Dropdown. This lists all form fields across your layers. Use it for precision:
"What is the average #Flow_Rate in @Fire_Hydrants?"
The mentions system uses keyboard navigation (↑/↓ arrows, Enter to select, Escape to dismiss).
4. Understanding AI Responses
Response Types
Answers may appear as a count, a value, a table or a map preview, depending on the question and available data.
| You want to know… | Example | What to check |
|---|---|---|
| How many records match | “Count inspections where Condition is Poor” | Layer, condition and total |
| A total or average | “Average Height in Inspection sites” | Field, units and missing values |
| A breakdown | “Group inspections by Condition” | Groups versus records analysed |
| The highest or lowest value | “Highest Height in Inspection sites” | Units and any tied records |
| Where results are | “Sites within 100 metres of Maintenance routes” | Distance, source layers and map |
| Work during a period | “Overdue inspections due in October 2026” | Date field, period and work status |
Reading grouped counts
A grouped-count answer separates output groups from the source records analysed. For example, 3 Condition groups · 39 Inspection sites analysed accompanies a Condition/Count table: Fair 13, Good 13, Poor 13, Total 39. The three groups are not three source sites.
A verified complete total is shown when available. If the source total is unverified or results are partial, read the limitation and do not treat the available groups as complete coverage. This presentation supports English and Arabic and uses the same grouped-table structure in chat and report previews. See the worked example.
Query Scope: Entire Layers by Default
Unless you explicitly request a smaller scope, Geova AI evaluates the complete referenced layer or project dataset available to your workspace. It does not limit a count, group, filter, or spatial analysis to features currently visible on the map simply because the map is zoomed in.
- Use
@Layerand#Columnmentions to identify the intended complete layer and field. - For a subset, name exact field values or a saved boundary and inspect the returned scope.
- Do not assume “the selected features” or “on screen” alone applies the selection or map extent to an answer.
- A mini-map is a result preview; it does not change the analytical scope by itself.
Always check the response scope, record count, SQL, and provenance before saving or exporting a result.
SQL Transparency
Many tabular responses include a collapsible "View SQL Query" section showing the filter or query logic that was executed:
SELECT * FROM "Fire_Hydrants" WHERE "Condition" = 'Poor'Use this section to review what was executed and confirm that the filter matches your intent.
Mini-Map
When results contain geographic features, an inline mini-map renders directly in the chat bubble, showing the spatial distribution of results. Below the map:
- Feature count indicator
- "Open Full Map" button to view results on the main map
What Geova AI Can Help With
Geova AI is strongest when your project has clear layer names, field names, and Data Dictionary entries. It can assist with:
| Category | Typical Use |
|---|---|
| Counts, filters, and summaries | Count records, filter by status, group by category, calculate totals or averages |
| Proximity analysis | Find nearest features, build buffers, compare features inside or outside a distance |
| Overlay and joins | Join points to zones, clip features by boundary, compare two layers |
| Service areas and access | Estimate catchments or reachable areas when the analysis template is available |
| Planning support | Rank candidate sites, identify gaps, compare scenarios, prepare exportable result layers |
| Reports and exports | Generate result tables, map previews, GeoJSON/CSV/KML exports, or PDF summaries |
AI results are decision support, not a replacement for professional review. For engineering, regulatory, safety, or legal work, confirm the layer schema, coordinate system, and generated filters before acting on the output.
Field Intelligence and Evidence Grades
Geova can run deterministic, read-only field checks without calling the language model. A successfully validated technical answer still consumes the same single workspace AI point as any other answer path. Try requests such as:
- “Check data quality for this layer.”
- “Show required-field completeness for the visible layers.”
- “Summarize conflicts in this layer.”
- “Which Inspection sites fall within the North service zone in Service zones?”
- “Create an inspection handover report.”
These checks use the active form, Smart Logic, current lexicons/domains, geometry validity, field types, and synchronization state. They do not change records.
Name the required layers explicitly. Ordinary analysis can include records outside the current map screen. For a smaller set, specify exact field conditions or a saved boundary and verify the returned records; map visibility and a current selection are not sufficient proof of the answer’s scope.
The same full-layer rule applies to short questions such as “show damaged roads” or “count roads by district.” Geova does not silently reduce these answers to the current map extent.
Evidence badges mean:
| Badge | Meaning |
|---|---|
| Decision-grade | Exact and complete deterministic evidence met the configured freshness, warning, and exclusion thresholds |
| Review required | The result is exact but has a warning, freshness concern, or excluded-data concern |
| Advisory only | The result is sampled, approximate, fallback, unknown, incomplete, or failed a validation gate |
The grade describes the data path, not professional approval. Always review safety-, engineering-, legal-, or regulatory-critical results.
Focused Form Co-Pilot and Controlled Field Actions
When Geova has context for one focused feature and form field, it can suggest a value from that field's current Lexicon or fixed options. It does not invent a new controlled value when the available choices are unknown, ambiguous, hidden by Smart Logic, or incompatible with the form.
A supported suggestion remains a draft until you review a one-field card showing the current and proposed values, source, confidence, and relevant rule effects. Apply requires explicit confirmation and rechecks the current record, form rules, domain version, project, workspace, role, and ownership. Undo is offered only while the applied value and record revision are still safe to reverse.
- Viewer and other read-only roles cannot apply a suggestion.
- Resolve unsaved edits in an already-open form before applying an AI field action.
- A hosted offline field action is blocked unless the device has valid offline authorization. Normal manual offline drafts continue to work.
- Bulk field mutation is not enabled; each proposal affects one feature and one field.
Area-Scoped Questions: Field or Boundary Layer
Be explicit about whether a place is a stored field value or a map boundary:
| Your intention | How to ask |
|---|---|
| Match a stored city value | “Count Fire Hydrants where City equals Toronto” |
| Use a geographic boundary | “Count Fire Hydrants inside the Toronto polygon in City boundaries” |
| Both kinds of data exist | Name the field or boundary layer you want to use |
| Neither exists | Add the required field or boundary data before asking for that comparison |
Read the answer's filters and scope. A city name alone does not establish which of these two questions you intended.
When Geova Needs More Detail
If Geova AI cannot safely understand the question, it should not guess. It shows a clarification card with action chips such as:
| Situation | What Geova Shows | Useful Next Action |
|---|---|---|
| Unclear question | A short note explaining that it needs a map, layer, field, or report question | Tap Explore Data, or add an @Layer / #Field mention |
| Multiple possible layers | A prompt explaining that more than one layer could match | Tap Choose Layer or Show Layers |
| No table records matched | What was checked and why the filter may be too narrow | Tap Broaden Search or check field values |
| No spatial features matched | A note about overlap, distance, filters, or coordinate systems | Tap Guided Pass or Check Layers |
| Fallback method used | A warning that the result used an estimated method | Review the notes before using the result operationally |
Clarifications, system failures, and unconfirmed empty results are not billed as useful answers.
One Point per Validated Technical Answer
A validated technical answer consumes one workspace AI credit (point), whether Geova used a quick deterministic check, a spatial analysis, or model assistance. Internal processing and retries for the same answer do not add another credit. Submitting a new question starts a separate request.
An exact answer of zero can still be a valid paid answer when Geova verified the complete requested scope. A clarification, failed operation, conversational reply, or unconfirmed cached/fallback result is not a validated technical answer.
For spatial analyses, recovery chips are matched to the operation:
| Analysis Type | What Geova Checks Next |
|---|---|
| Buffers and service areas | Distance/range, units, facility layer, routing fallback, and target overlap |
| Spatial joins, clips, overlays, and comparisons | Layer overlap, containment direction, filters, geometry validity, and CRS alignment |
| Nearest, travel, and OD-style analysis | Origin/reference point, destination layer, GPS, valid coordinates, and route fallback |
| Density, clustering, and heatmaps | Point layer suitability, feature count, layer extent, grid size, or cluster count |
| IDW/interpolation | Numeric field selection, point sample density, boundary mask, and grid resolution |
| Gap, suitability, and optimal placement | Service range, boundary layer, exclusion layers, and whether a gap actually exists |
| Catchments/Voronoi and load simulations | Facility count, duplicate locations, boundary mask, and load assumptions |
| Carbon and polygon measurements | Polygon geometry, area validity, and the calculation factor |
5. AI Prompt Cookbook
Use these examples as starting points. Replace layer and field names with your own, or use @Layer and #Column mentions so Geova AI can target the right data.
| Goal | Example Prompt | What to Review |
|---|---|---|
| Count incomplete work | "Count records in @Hydrant_Inspections where #Status is not Complete" | Confirm the status field values match your form options |
| Find high-priority repairs | "Show @Streetlights where #Condition is Poor or #Repair_Priority is High" | Check whether the logic should be OR or AND |
| Group records by category | "Group @Road_Defects by #Defect_Type and show counts by district" | Confirm district values are present and standardized |
| Find missing evidence | "List @Damage_Assessments where the photo field is empty" | Review whether the photo field is single photo or gallery |
| Use current location | "Find the 10 nearest hydrants to my current location" | Confirm device GPS accuracy before using the result |
| Check distance from another layer | "Find assets within 100 meters of @Road_Closures" | Confirm the distance unit and source layer |
| Join points to zones | "Assign each @Inspection_Point to the containing @Management_Zone and summarize counts by zone" | Review the zone layer and any points outside all zones |
| Compare two areas | "Compare open repairs in Zone A and Zone B by priority" | Confirm zone names and priority values |
| Prepare a result layer | "Create a result layer for poor-condition assets inspected this month" | Preview the map result before tapping Add to Mapplex |
| Build an export | "Export all incomplete inspections as CSV with asset ID, condition, inspector, and date" | Check selected columns and date range |
| Screen candidate sites | "Rank candidate parcels by area, zoning type, and distance to main roads" | Confirm required fields/layers exist and review ranking assumptions |
| Find service gaps | "Identify neighborhoods more than 500 meters from a clinic" | Confirm clinic and neighborhood layers, and whether straight-line distance is acceptable |
Prompting Tips
- Name the layer and field whenever possible.
- Use the same wording that appears in your form choices, such as
Good,Fair,Poor,Open, orComplete. - Ask one operational question at a time, then refine the result.
- Review the SQL, result table, and map preview before exporting or saving a layer.
- For location-sensitive prompts, wait for GPS accuracy to stabilize first.
6. Action Buttons
Expand Result tools & export beneath a response to find its available actions. A count or explanatory answer does not necessarily include downloadable features, a map, or a PDF. The record count is an indicator, not a button:
| Button | Action |
|---|---|
| 📊 Records | Shows the total record count |
| 🔍 Go Deeper | Re-runs the query through a structured analysis template when available |
| Generates a PDF report of the results | |
| ⬇️ GeoJSON | Exports results as a GeoJSON file |
| 📋 CSV | Exports results as a CSV spreadsheet |
| 🌍 KML | Exports results as KML for Google Earth |
| 💾 Add to Mapplex | Saves AI results as a permanent layer in your project |
| 🗺️ Preview | Shows results on the main map with AI preview overlay |
| 🔗 Dissolve | Merges overlapping result polygons where this action is available |
Committing Results as a Layer
- Review the response's scope, source, record count, and map preview.
- Tap Add to Mapplex while the spatial result is still available.
- Enter a layer name in Add AI Layer.
- Review Confirm AI Layer Creation, including the destination project and number of features, then confirm.
- Wait for the success message and verify the layer in Layers and Map.
Mapplex checks permission and the current project/workspace before creation. If the same result action was already applied, it can report that the layer exists instead of creating a duplicate. Preview and Open Full Map display temporary results; they do not perform this save.
Exporting Results
Choose GeoJSON, CSV, or KML when that response has stored result data. Mapplex rechecks export access, then prepares the file. Check the exported count and open the file in the receiving tool. Exporting an AI result uses that result's scope; it is not the same as Entire Project export in Data & Sync.
Keep useful results: Chat text can survive longer than its temporary result data. If Mapplex says the result is no longer available, run the analysis again; a new question is a separate request under the credit rules. Save an approved result as a layer or export it while available.
Worked example: poor-condition assets
Assume your project contains a Hydrants layer with a Condition dropdown.
- Use mentions to ask Show records in @Hydrants where #Condition is Poor.
- Check that the answer names the intended layer, matches the Poor option, and covers the intended dataset.
- Inspect the count, evidence, SQL when available, and map preview. A count of zero may be a verified answer; do not automatically assume a failure.
- Use Preview to inspect locations. Use CSV for a review list, or Add to Mapplex and its confirmation steps for a saved result layer.
- Reopen the file or saved layer and compare its count with the result.
If you need a smaller area, name a saved boundary and check that scope in the answer. Benefit: You can produce a focused inspection list without changing the source assets.
7. Session Management
Per-Project Sessions
Each project has its own chat session. Switching projects switches the conversation. When local persistence is enabled, recent history is kept on that device, subject to retention and storage limits; it is not a permanent project archive or a promise of cross-device history.
Private Chat
Tap the header's lock control, Private chat — do not save history on this device, to start a separate private conversation. Switching modes stops an active request and clears the displayed conversation. Earlier saved history is kept separately; new private messages are not saved on the device. Switching private mode off restores the earlier saved history and discards the private conversation.
Private mode changes local history storage. Requests still use the Geova Cloud and the processing described under Privacy & Data Safety. It does not make the AI request offline or delete provider-side records. Check the lock state when starting another session. Use Clear Chat History separately if you intend to remove the project's saved local history.
Clearing History
Tap Clear Chat History in the header and confirm to clear the current project's conversation and its saved history on this device. This does not delete a project layer already created from a result or a file you already exported.
Credit System
- Each useful AI answer costs 1 credit
- Quick, deterministic, deep, and model-backed answer paths have the same one-credit price
- Clarification questions and system failures are not billed
- Credit usage is displayed in the input area
8. Review a multi-step answer
A question can combine several steps, such as filtering poor-condition assets, finding those near a route and counting them by zone. Before using the result:
- Check the project, layers and field values named in the answer.
- Confirm distances, units and date ranges.
- Compare the result count and map with the source records.
- Read any warning about missing data or an estimated method.
- Save or export only when the result answers the question you intended.
If a calculation fails, retry after resolving the reported problem. An incomplete operation is not a successful empty result. Chat needs an internet connection even when Mapplex performs calculations on the device.
Review a Work proposal
In an enabled Teams/Enterprise workspace, chat can propose creating asset work orders or assigning existing work orders. Owner/Admin roles can assign members; Editors can create work where permitted but cannot assign it. Collectors can review their permitted work; a chat request does not give them manager permissions.
- Ask for the intended action, naming the assets or work orders. For new work, include its type, due date and priority.
- Review the proposal's action, number of orders, listed records and proposed assignee.
- Choose Confirm to proceed, or Cancel to leave the work unchanged. Confirmation requires connectivity and checks current access and records again.
- Read the completion message, then check Work schedule. Creation and assignment can finish separately; if only some orders succeed, inspect what remains before making another request.
If the records, project or account changed since the proposal, ask again and review the new proposal. A proposed action is not a completed assignment. See Work Schedule and My work.
9. Error Handling
| Error Type | What You See | What to Do |
|---|---|---|
| No Data Found | Recovery note with action chips | Broaden search, check fields/layers, or run a guided pass |
| Auth Error | Red bubble with sign-in warning | Sign in again or refresh your session |
| Out of Credits | Red bubble with credit warning | Recharge workspace credits or ask an admin to upgrade |
| Network Error | Red bubble with retry button | Tap "Try Again" |
Voice Assistance During Capture
Fill by voice is available inside Quick Capture for enabled workspaces. It fills the current draft through a reviewed suggestion. Use Geova AI Chat for questions about project data. A valid voice suggestion delivered for review costs one workspace point; applying it and saving the record do not add another voice charge.
10. Privacy & Data Safety
- Geova Cloud processing — Your question and the relevant project structure are sent to Geova AI. This can include layer and field names, labels, units, configured choices and limited sample values.
- Limited follow-up context — Counts and totals can be used to explain an answer. If a term needs clarification, a limited set of values from the relevant field may also be used.
- Record lists stay in Mapplex — For project-data chat analysis, Mapplex uses the records it holds to display result rows. Full records, record identifiers and geometries are not sent to the AI service as result rows. This does not prevent information you type into a question, or information present in a sample value, from being sent.
- Location-based questions — Review the location and scope when asking “nearest to me.” A request that needs an external routing service may send the coordinates needed for that route; check the processing information for the tool you use.
- Your access still applies — Questions and confirmed Work actions stay subject to workspace, project and role permissions. Check the project shown before proceeding.
- Private chat controls device history — It does not make requests offline or change cloud retention. Voice entry and Planner have their own processing notices; read them before submitting.
- Choose appropriate content — Use information your organization permits for AI processing. Review answers before relying on them for field decisions.
See the Mapplex Privacy Policy for provider, retention, deletion, and data-transfer details. Account deletion requests are available at geova.net/account-deletion.
Next: Proceed to Module 07: Team & Collaboration to manage team members and monitor field performance.