Beta Feature: Template Recommendations is currently in beta and available for Google Slides templates. Contact your Technical Account Manager for access.
The best templates are refined over time as you learn what your end users actually need. Template Recommendations gives you a way to do that continuously: it reads the customizations your end users make to presentations generated from a template, finds the ones that repeat, and turns them into specific recommended changes to the template. Each recommendation shows how many presentations it came from, how many end users made the edit, and which presentations it is based on, so you can check the evidence before acting.
The result is a template that keeps improving with use. The more presentations your team generates, the clearer the patterns become, and the less time you spend guessing what to change.
Recommendations appear in a Recommendations tab on the template's Setup page. Any admin with edit access to the template can run an analysis and act on the results. End users never see recommendations.
The Analysis Process
An analysis compares each presentation as it was generated with the presentation as it is now, then looks for patterns across all of them. Matik combines three methods to do this reliably. Matik's own code handles everything that can be calculated: which presentations are included, what changed on each slide, which template slide each edit belongs to, and every count and percentage. AI handles what requires reading and judgment: describing an edit in plain language and recognizing that the same kind of edit repeats across presentations. Rubrics and evaluations define what a good recommendation looks like, verify each AI proposal against the presentations, and test the analysis against known cases.
The process starts before any analysis runs. When a Google Slides presentation finishes generating, Matik records what it looked like at that moment. When you run an analysis, the Recommendations tab shows each stage as it is reached:
- Queued. The analysis is waiting to start. Matik confirms there are enough presentations to analyze.
- Snapshotting. Matik reads the current version of each presentation generated from the template.
- Diffing & anchoring. Matik compares each presentation to the snapshot taken at generation. The differences are the edits your end users made. Each edit is matched to the template slide it came from.
- Categorizing changes. For each edited presentation, Matik's AI writes a description of what changed.
- Synthesizing recommendations. Matik's AI reads all of those descriptions together and proposes recommendations where the same kind of edit repeats across presentations.
-
Grading and ranking. Each AI proposal is verified against the presentations before it becomes a recommendation:
- Citations are verified. Every presentation a proposal cites is checked against the actual set of analyzed presentations. A citation that does not match a real presentation is removed, and a proposal with no valid citations is dropped.
- Counts are calculated. The presentation count, the distinct end-user count, and the percentage on every recommendation are calculated from the verified citations. Any number in the recommendation's text is cross-checked against those values, and a mismatch is recorded.
- Confidence is capped. The AI proposes a confidence tier. The numeric rules in Confidence Tiers act as a ceiling and can only lower it.
- Weak proposals are dropped. Anything below the evidence floor does not appear.
- Details are read from the presentations. The destination of a moved slide, the library slide inserted, the text end users keep adding, and the Dynamic Content that returned no data are all read directly from the cited presentations' edits.
- Slide references are validated. A proposal that points at a slide no longer in the template is dropped rather than reassigned.
- Duplicates are dropped. When two proposals describe the same change to the same slide and element, Matik keeps the one backed by more presentations. A vague Other proposal is dropped when a specific one exists for the same slide.
Tracked Data
Analyzable Presentations
The Recommendations tab counts a presentation as analyzable when it meets all of the following:
- Generated from this template. Each analysis is scoped to one template.
- Google Slides. PowerPoint and email templates are not included.
- Non-test generations. Test generations and previews from the template editor are excluded.
- Generated by admins or end users. Both are included.
- Generated within the analysis window. The window ends when you run the analysis and reaches back at most 90 days. It never reaches back further than the date Matik started collecting edits, which the tab shows as Collecting edits since [date].
- Narrative Agent presentations are counted but not analyzed. The agent edits the presentation after generation, so its work cannot be separated from end-user edits. These presentations appear in Decks generated and are skipped as No snapshot.
Presentations that have not been edited still count. An untouched presentation is evidence that a pattern is not universal, which keeps a recommendation from overstating how common a change is.
Captured Data
At generation, Matik records the presentation's slide content as Google Slides reports it: the text in each element, its position, size, and formatting, and which template slide it came from. Matik also records a thumbnail image of each slide and notes which Dynamic Content on each slide returned no data.
When you run an analysis, Matik reads the current version of each presentation the same way. Because a presentation lives in the end user's Google Drive, Matik reads it using that end user's own Google connection to Matik. This is why an analysis runs only when an admin asks for one and never on a schedule.
Comparing the two versions tells Matik what changed: text rewritten, values corrected, formatting changed, elements moved, resized, added, or deleted, and slides removed, moved, or inserted. Matik sees the current state of the presentation, not the history of edits, so it cannot tell when an edit was made or who made it. Evidence is attributed to the end user who generated the presentation.
Data Shared with Matik's AI
Matik's AI receives, for each edited presentation, the list of changes on each changed slide (text before and after, formatting and layout changes, elements added or removed), the template's slide descriptions and Dynamic Content names, and before-and-after images of the changed slides. It does not receive end user names, email addresses, or presentation names. For the AI partners Matik uses, see Managing AI Features.
Not Tracked
- How presentations performed after they were shared, such as email engagement, CRM outcomes, or call recordings.
- Edits made in a copy of a generated presentation. Matik reads the presentation it generated, so edits in a duplicate or a downloaded file are not visible.
- Changes an end user makes before generation, other than which Slide Library slides they chose.
- Comparisons between templates. Every analysis covers one template.
Running an Analysis
- Open the template and go to the Recommendations tab next to Outline on the Setup page.
- Click Run analysis. If a report already exists, the button reads Run analysis again.
- Confirm in the Run template analysis? dialog. Processing usually takes up to about 15 minutes. Large templates can take longer.
You can leave the page. Matik emails you when the analysis finishes, whether it produced recommendations, found none, or failed. The link in the email always opens the template's current recommendations.
The tab shows how many analyzable presentations the template has so far. An analysis needs at least 5. Until then the button is unavailable and the tab reads Recommendations are not available yet.
Your organization can run two analyses at once across all templates, and one per template. If both slots are in use, the tab names the templates being analyzed and you can start this one when one of them finishes.
If the template already has recommendations, they stay visible and fully actionable while the new analysis runs. When it finishes, the list refreshes in place. If the analysis fails, the existing recommendations stay as they were, with a notice explaining what went wrong.
The Recommendations Report
Run Summary
The top of the tab shows when the analysis ran, who requested it, and the window it covered, followed by three counts:
- Decks generated: presentations generated from the template within the window.
- Edited by an end user: analyzed presentations where Matik found at least one change.
- Analyzed by Matik: presentations Matik was able to compare. A skipped count next to it breaks down why any were left out: No snapshot, Could not map, or Analysis error. See Troubleshooting.
Below the counts, What end users did with this template is a short AI-written summary of the edits across all analyzed presentations.
If the analysis ran and found no repeating pattern, the tab reads No recommendations in this report. That is a normal result. The summary counts still show what was analyzed.
Recommendation Details
Recommendations are listed in priority order. Each one shows:
- Headline and reason: the proposed change in one line, with a one-sentence rationale.
- Type: which kind of change it is. See Recommendation Types.
- Slide: which template slide the recommendation points at, with a thumbnail of that slide as it is in the template now. Recommendations that apply to the template as a whole read Whole Template.
- Confidence: High confidence, Medium confidence, or Early signal. See Confidence Tiers.
- Evidence line: how many analyzed presentations show the pattern, how many distinct end users generated them, and the percentage of analyzed presentations. For example, 6 of 14 analyzed decks · 4 distinct end users · 42.9% of analyzed presentations.
- Tags: New for a recommendation the latest analysis found for the first time, and Sent to Matik AI with a date once it has been handed to Matik AI.
Evidence
Click View more details on a recommendation to see Why this is recommended, a longer explanation, and Presentations this is based on, the list of presentations the recommendation cites. Each entry shows the presentation's name, who generated it, and when. Presentations you have access to are linked. Presentations you cannot access are listed with a No access tag and are not linked, and presentations that have since been deleted show Deleted presentation. You can search the list and, when more than one end user is cited, filter it by end user.
The number of cited presentations always matches the count on the evidence line.
Recommendation Types
Matik produces eleven kinds of recommendation. Each one is triggered by a specific pattern in end-user edits.
| Type | Recommended Action | Signal |
|---|---|---|
| Add a slide condition | Skip the slide automatically when its data comes back empty. | End users delete a slide whose Dynamic Content returned no data when the presentation was generated. |
| Check data | Look into a value, chart, or table. | End users correct the same generated number, chart, or table in the same way. |
| Fix formatting | Make a formatting correction in the template. | End users make the same formatting change, such as shrinking a font or widening a box, to the same element. |
| Move a slide | Put the slide where end users move it. | End users move the same slide to the same position in the presentation. |
| Rework an element | Change or remove an element on a slide. | End users delete the same element while keeping the slide. |
| Remove a slide | Review whether the slide belongs in the template. | End users delete the same slide. |
| Add content | Add text to the template. | End users add the same text in the same place across presentations. |
| Add a library slide | Make a Slide Library slide part of the template. | End users keep inserting the same Slide Library slide, often at the same position. |
| Automate content | Generate the text from Dynamic Content instead. | End users keep manually populating text that could be automated. The text differs from one presentation to the next, and it could be generated from the presentation's own data or from a connected data source. |
| Tune AI prompt | Adjust the prompt behind an Insights Dynamic Content. | End users rewrite the output of the same AI-generated text in the same direction. |
| Other | Review a pattern Matik could not classify. | A repeated edit that does not fit any of the types above. |
When Matik Does Not Recommend
A recommendation only appears when an admin could reasonably make the change in Matik. Matik does not produce a recommendation for:
- Presenter notes and instructions. Elements that exist to be read and deleted, such as a "Note to CSM" box, or that tell the presenter to do something outside the presentation. End users deleting these is the template working as intended.
- Content a person decided. Owners, assignees, target dates, commitments, and next steps that name a person or a date. This information does not exist in any data source Matik could pull from.
- Scattered edits. Many presentations edited, but no two sharing the same change to the same slide or element. That a template gets edited a lot is not a recommendation.
Confidence Tiers
Matik's AI proposes a confidence tier for each recommendation based on how consistent the edits are. Matik then applies numeric rules as a ceiling. The rules can lower a tier but never raise one.
| Tier | Every condition must hold |
|---|---|
| High confidence | At least 60% of analyzed presentations show the pattern. At least 3 distinct end users generated them. At least 10 presentations were analyzed. The edits are consistent. |
| Medium confidence | At least 30% of analyzed presentations show the pattern. At least 2 distinct end users generated them. The edits are consistent, or the same action on the same target with content that varies. |
| Early signal | Anything that clears the floor below but does not meet Medium. Treat it as worth a look, not a conclusion. |
Two rules apply on top of the table:
- The floor. Matik does not show a recommendation at all if fewer than 3 presentations show the pattern, if fewer than 20% of analyzed presentations show it, or if it comes from a single end user and fewer than 50% of analyzed presentations show it.
- Hard caps. A pattern from one end user is never higher than Early signal, because one person's habit is not a template problem. When fewer than 10 presentations were analyzed, nothing is higher than Medium confidence.
The percentage is always out of all analyzed presentations, including ones that were not edited. Confidence does not consider how recent the edits are.
Expect mostly Medium confidence and Early signal in the first few analyses. High confidence needs both a large sample and broad agreement among end users.
Priority order. Recommendations are sorted by confidence weight multiplied by the percentage of presentations showing the pattern, with High confidence weighted 3, Medium confidence 2, and Early signal 1. Ties go to one-step fixes before changes that need judgment, then to slide-level changes over template-level ones, then to earlier slides. Active recommendations are listed first, then any marked Worth another look, then Done and Dismissed.
Changes Between Analyses
A template has one set of recommendations that refreshes in place each time you run an analysis. There is no list of past reports.
- Each analysis recounts from scratch. Counts never accumulate across analyses. Each run looks at the presentations in its window and reports what it finds.
- A recommendation keeps its identity. The same action on the same slide, and on the same element or Dynamic Content where one applies, is the same recommendation from one analysis to the next, even if the wording, counts, or confidence change. Matik identifies template slides by their stable Google Slides identity, so this survives reordering and re-syncing the template.
- Recommendations drop out, appear, and update. A pattern the latest analysis no longer finds disappears. A pattern found for the first time appears with a New tag. A pattern that persists picks up the latest counts and confidence.
- Dismissed stays dismissed. When you dismiss a recommendation, the dismissal attaches to the pattern. It arrives already dismissed in later analyses instead of coming back as new.
- Worth another look. If a dismissed pattern comes back with higher confidence than it had when you dismissed it, it stays dismissed but is highlighted with Worth another look. This signal strengthened from [previous tier]. Click Keep dismissed to acknowledge the new confidence and remove the highlight, or Help me do this to make it active again and open Matik AI.
- Done comes back only on new evidence. A recommendation you marked done returns to active only when the latest analysis finds the pattern in a presentation generated after you marked it done. Presentations from before your fix are not evidence that the fix failed. If it does come back, that tells you the fix did not take.
- A slide that leaves the template takes its recommendations with it. If you delete a slide from the template, recommendations pointing at it read Slide no longer in template until the next analysis, after which they drop out.
Note: Matik does not check whether the template already contains the recommended change. If you make a fix and run an analysis before your end users have generated new presentations, the same recommendation appears again with the same evidence. Mark it done so Matik knows to watch for new presentations only.
Recommendation Actions
Each active recommendation offers three actions. Mark done and Dismiss are shared: every admin with edit access to the template sees the same state, and each action records who took it. Matik AI conversations are personal to the admin who started them.
- Help me do this opens Matik AI with the recommendation and template already in context. For changes to Matik settings, such as adding a Slide Condition, editing an Insights Dynamic Content prompt, adjusting a data query, or tagging an element to Dynamic Content, Matik AI can make the change once you confirm it. For changes to the Google Slides file itself, Matik AI walks you through the edit and reminds you to sync the template in Matik afterward. Once started, the button reads Resume and the recommendation shows a Sent to Matik AI tag, for you only. Other admins see Help me do this and get their own conversation. Matik AI never marks a recommendation done for you.
- Mark done records that you made the change, whether through Matik AI or on your own. The recommendation collapses to a muted row with a Done tag.
- Dismiss records that the recommendation is not worth acting on. The recommendation collapses to a muted row with a Dismissed tag and stays dismissed in later analyses.
Done and dismissed recommendations stay in the list so you can see what has been handled. Click Restore on either to make it active again. Every action is reversible.
Best Practices
- Sync the template after editing the Google Slides file. New presentations always generate from the current file, but the analysis matches edits against the slides Matik synced last. If you change the file without syncing, edits on new or moved slides may not map to the template correctly.
- Ask end users to edit the generated presentation in place. Matik reads the presentation it generated. Edits made in a copy or a downloaded file are invisible to the analysis.
- Wait for new presentations before running again. Running an analysis immediately after a fix recounts the same presentations and shows the same recommendations. Run again once your end users have generated new presentations from the updated template.
Limitations
- Google Slides templates only. PowerPoint and email templates do not show the Recommendations tab.
- Narrative Agent presentations are not analyzed. They count toward the number of presentations generated but are skipped in every analysis.
- The window is at most 90 days. An analysis covers presentations generated in the last 90 days, and no earlier than the Collecting edits since date shown on the tab.
- One analysis per template at a time, and two per organization. An analysis in progress cannot be cancelled.
- A new analysis replaces the previous one. There is no history of past analyses.
Troubleshooting
| Message | Cause | Resolution |
|---|---|---|
| The Recommendations tab does not appear | The template is not Google Slides, you do not have edit access to it, or your organization is not enrolled in the beta. | Check the template type and your access. Contact your Technical Account Manager about beta enrollment. |
| Recommendations are not available yet / [n] of 5 analyzable presentations so far | Fewer than 5 presentations have been generated from the template since Matik started collecting edits. | Wait until more presentations have been generated, then run the analysis. |
| [n] reports are already generating for your organization | Your organization is already running the maximum of two analyses. | Wait for one of the named templates to finish, then run this one. |
| Only [n] of 5 analyzable presentations were available for this run | Some presentations counted toward the minimum could not be analyzed, for example Narrative Agent presentations or presentations regenerated after their snapshot was taken. This can also happen partway through a run if too many presentations are skipped. | Wait until more presentations have been generated, then run again. |
| No snapshot (skipped presentation) | The presentation was generated through the Narrative Agent, its creator's Google connection was unavailable when it was generated, Matik's snapshot at generation failed, or it was regenerated after its snapshot was taken. | No action needed. New presentations are captured automatically. |
| Could not map (skipped presentation) | None of the presentation's slides match a slide in the template as it is now, usually because the template's slides were all replaced since the presentation was generated. | No action needed. Presentations generated from the current template are mapped normally. |
| Analysis error (skipped presentation) | Matik could not read the presentation from Google Drive, or the AI step for that presentation timed out. If it is a Drive issue, the end user's Google connection to Matik may be disconnected, or the file may have been deleted or its sharing changed. | Run the analysis again. If the same presentations keep failing, ask the end user to reconnect Google in Matik. |
| Matik AI could not complete the analysis | The AI step failed during the run. | Run the analysis again. If it fails repeatedly, contact Matik Support. |
| Matik could not start the next analysis step | A step in the run could not be started. | Run the analysis again. |
| Matik could not complete the analysis because of an unexpected error | The run hit an unexpected error. | Run the analysis again. If it fails repeatedly, contact Matik Support. |
| The analysis stopped making progress | The run stalled and was stopped. | Run the analysis again. |
| Slide no longer in template (on a recommendation) | The slide the recommendation points at was removed from the template after the analysis ran. | No action needed. The recommendation drops out on the next analysis. |
| The report from this link has been replaced by a newer analysis | You opened an email link to an analysis that a newer one has since replaced. | No action needed. The page shows the current recommendations. |
Comments
0 comments
Please sign in to leave a comment.