AI Chat

AI Chat is Genspark's chat for text. You ask, and it answers in the conversation. When search is on it cites what it used. It reads the files you attach, and it drafts or rewrites text you paste in.

It has its own model picker and its own 5-hour limit. Neither is shared with Super Agent or with Free Quota. Use it for the thinking and the wording, then hand the result to an agent when you want something built.

What is AI Chat?

AI Chat is one conversation with one model behind it. You pick the model before the first message and keep it for the length of the chat. It answers questions, searches the web when search is on, reads what you attach, and writes or edits text.

AI Chat can reach three things: the web through Search Web, the files you attach, and Genspark's own product documentation, which it can read whether or not search is on. It doesn't read your Hub files, and it has no memory of your earlier chats.

What you can do

Six jobs cover most of what people bring to AI Chat. Pick the model first, because it locks after your first message (Quick start). Then paste a prompt with your own text under it, or attach the file the prompt names. All thirteen recipes are at the end, under Recipes.

Two sections right after Quick start apply to every job: the AI Chat 5-hour limit and which models cost credits.

Where everything is

The Genspark home page composer is another way in, but it can start Super Agent, so check which agent the box names before you type a long prompt into it.

Quick start

Before you start: a Genspark account and a question. Pick the model before you type, because the picker locks after the first message.

1. Open AI Chat from the sidebar Agents menu (Tools → AI Chat), from AI Chat in the home page quick-start bar, or from All Agents → Content Creation → AI Chat. In the app, open the AI Chat screen and tap New Chat. A phone browser uses the web page.
2. Open the model picker in the composer and choose a model. A fast model is selected by default, and a new chat opens on the model you last picked in this browser if it's still offered. If you aren't sure, keep the default. Whether a model uses credits depends on your plan (Which models cost credits, by plan).
3. Type your question and send it. From then on the picker's model list is locked: opening it reads "You need to start a new chat to switch models" with a New Chat button. Search Web and Personalization below the list still work.
4. Follow up in the same chat.
5. To change the model, open New Chat and pick before you type again. The old chat keeps its model and its history.
6. Turn on Search Web or Personalization in the picker dropdown when the job needs them. Search Web can be turned on later in the chat too. Once search has run, it can't be turned off, except with Mixture-of-Agents. Personalization appears once you've filled in a profile.

Two habits help. Ask for the shape you want, such as "five bullets" or "a table". And start a new chat when the topic changes, since a chat carries everything said before it.

AI Chat 5-hour limit

AI Chat has a usage limit of its own: a fixed 5-hour window, per user, that starts with your first message that counts toward it. Free accounts, Enterprise seats and Team organizations billed by invoice don't have it. New chats don't reset it. When the window ends, the counter starts again with your next message.

The window covers AI Chat only, and it isn't part of Free Quota. Free Quota has its own weekly limits, and AI Chat draws on neither. Other agents keep working while AI Chat is limited.

Depending on your plan, AI Chat tells you where you stand in one of four ways. On Plus and Pro from September 18, 2026, a strip above the composer shows the first three while a core model is selected, and you can close it with ✕. On plans that stop at the limit, the fourth comes as a reply in the chat and a Session Limit Reached dialog. Core models and flagship models are the two picker groups on those newer plans, explained in Core models and flagship models.

In the last message, the words 5-hour limit link to Membership Plans, which carries the plan-level wording. Which plans block and which keep running is in Which models cost credits, by plan.

Moving from Plus to Pro during a window resets it once, so Pro's larger window starts clean. A downgrade keeps what you have left, capped at the new plan's limit.

Which models cost credits, by plan

Every message is billed one of two ways: free inside the window, or on credits. Which one applies depends on the model and on the plan. Your plan picks the section below. If your picker opens on a No credit cost section, you're on Plus or Pro from September 18, 2026. Free accounts, Plus and Pro from before that date, and Team and Enterprise seats see Recommended models followed by models grouped by maker.

Core models and flagship models

On Plus and Pro from September 18, 2026, the picker splits models in two. Core models cost no credits while the window has room. Flagship models always use credits, and they don't use up the window. Other plans don't have this split. See their sections below.

On those plans the core section is headed "No credit cost" while the window has room and "Limit reached" when it's spent. It shows four models first, and More opens the rest. Flagship models follow under Flagship models, grouped by maker. Every row shows its cost as a multiplier such as "2×", and flagship rows also carry a Credits tag. Free accounts, Plus and Pro members who subscribed before September 18, 2026, and Team and Enterprise seats see Recommended first and then models grouped by maker, with no core and flagship split.

A Mixture-of-Agents message counts as core unless the mix includes a flagship model, and then the whole message bills credits.

Free accounts

Every message uses credits, and a free account gets 100 credits a day for it. There's no 5-hour limit on a free account. The picker shows Recommended and then models grouped by maker, and pricier models cost more credits per message. Some models are for members only: a free account sees them but can't run them, and on some accounts flagship models are locked behind Unlock top models.

Plus and Pro from September 18, 2026

If your subscription started on or after September 18, 2026, or you came back after a lapse of more than 14 days, core models are free inside the 5-hour window. When the window is spent, core models keep working on credits, so the chat keeps going until your credits run out. Then the composer says so, and buying credits lets you continue. Flagship models always use credits.

Pro has a higher limit than Plus, so its window holds more model spend.

Plus and Pro from before September 18, 2026

If your subscription started before September 18, 2026 and stayed active, every AI Chat model is free, flagship included, and the window blocks the chat when it's spent. A lapse of up to 14 days still counts as continuous.

That benefit runs until December 31, 2026. From January 1, 2027 the rules above apply. Membership Plans and Credits Guide carry the plan-level wording.

Team and Enterprise seats

Team and Enterprise seats keep every AI Chat model at no credit cost, whatever date the seat started. Team seats have the 5-hour limit, the same size as Plus, and stop at it. Enterprise seats, and Team organizations billed by invoice, have no AI Chat limit. The picker shows Recommended and then models grouped by maker. An Enterprise admin can limit which model vendors appear.

One wallet for every product

Credits come from one wallet that every Genspark product draws on, so a credit spent in AI Chat is a credit Super Agent can't use.

Next: Model picker covers the controls these labels sit in.

Model picker

There are no Standard or Ultra tiers in AI Chat. The picker has a search box and lists models by maker. On Plus and Pro from September 18, 2026, a No credit cost list of core models comes first and the flagship models follow under Flagship models. On every other plan a short Recommended list comes first. Mixture-of-Agents is listed under Genspark, near the bottom.

The search box at the top finds a model by name or by maker. On Plus and Pro from September 18, 2026, the list shows four core models first, and More opens the rest, flagship models included. The lineup changes often, so the picker is the list to trust. The models you see can vary by plan, region, organization settings and gradual rollouts.

The model is chosen before the first message. After that the model list is dimmed, and opening it reads "You need to start a new chat to switch models" with a New Chat button that opens a new chat in a new tab. A chat keeps its model to the end, unless that model is retired. Then the chat moves to its replacement.

Mixture-of-Agents

Mixture-of-Agents is listed under Genspark in the picker. Several models answer the same question and an aggregator writes one reply from their answers.

It's slower than a single model, because more than one model runs. Use it when the question is unclear or the stakes are high, and you want one answer that weighs several. The mix isn't shown in the picker. It differs with Search Web on or off and it changes over time. On Plus and Pro from September 18, 2026, it counts as a core model. On a free account every model in the mix uses credits. If you need to be sure of the cost, pick a single model.

Web search and sources

Search Web is a switch in the picker dropdown, described as "Retrieve from Internet". Turn it on before the first message or at any point later in the chat.

With Search Web on, replies put a [Source] link after the sentence or paragraph that relies on a search result, on every plan. How closely a model follows this varies, and Mixture-of-Agents replies don't carry the links. Open two or three of the links and check that each source says what the reply says.

Once search has run in a chat, you can't turn it off. The Search Web switch in the picker then reads "Search cannot be disabled after using search tools." Start a New Chat if you want a chat without it. Mixture-of-Agents is the exception: its search switch stays free.

If a model can't search, the Search Web switch is grayed out and stays off. Choose another model in a new chat.

Genspark's own product documentation is a separate tool, covered in Product documentation.

Files and images

AI Chat reads what you attach. AI Chat accepts PDF, EPUB, CSV, MD, TXT, HTML, JSON, XML, YAML, Word, Excel, PowerPoint, code files and images. The limit is 100 MB per file, and 80 images per conversation.

Google Docs, Sheets and Slides can't be attached as they are. Download the file as Word, PDF or Excel first, then attach that.

Large text files are read in parts, on demand.

Most models read images. With a model that can't, the file picker doesn't offer images, and pasting one shows "This model does not support images".

Personalization

Personalization is on by default. It adds what you've written in Settings → Personalization to each reply: your nickname, occupation, profile and chosen traits, plus your custom instructions on paid plans. Turn it off for a chat with the Personalization switch in the model picker, under Search Web. The switch reads "Apply your profile preferences" and appears once you've filled in your nickname, occupation, traits or profile.

Each chat remembers its own setting, and a new chat starts with the setting you used last in this browser. If you turn it off for one chat, switch it back on in the next if you want it. You edit the profile in Settings → Personalization, and it applies to every chat.

Personalization is a profile you set, not a memory of past chats. AI Chat has no memory system, and it doesn't read your Hub files.

Product documentation

Genspark's own product documentation is always available to AI Chat, whether or not Search Web is on. That's how it answers questions about Genspark products.

Long chats

AI Chat compacts older history automatically as a chat gets long. When it does, the chat shows "For better performance, previous chat history has been compacted." Older history is compacted to make room, so the chat keeps going.

If a chat still overflows, the reply stops with "Context Length Exceeded, Please open a new session" and a Create a new session button. That's the point to start a new chat. Asking for a summary earlier makes that easier (Keep a long conversation useful).

A new topic deserves a new chat. There's no ring or compress control in AI Chat: the context ring and the compress button belong to Super Agent, and they don't appear here.

Where AI Chat stops

AI Chat answers and writes. When the job needs a file, a computer or a picture, another product takes it. There's no button from AI Chat to any of them, so the handoff is a copy and a paste.

Shape a prompt turns an idea into a brief you can paste into another product.

Troubleshooting

Find your symptom in the table.

You hit the limit with credits left

The window and your credit balance are separate counters. The message in the composer shows which one stopped the chat. The four messages are in AI Chat 5-hour limit.

A reply that repeats itself

Start a New Chat with the same question. If it repeats again, pick another model before you type.

The chat won't load or finish

If something fails on Genspark's side, the reply shows "There was an error generating a response." with a Regenerate response button. Regenerate response first tries to recover the reply. If it has to run the request again, that run uses credits like any message. If it keeps failing, wait a few minutes and reload. The site won't load (524) and The app is slow, stuck, or won't open cover the rest.

The chat stopped with a context error

Start a new chat. To avoid this next time, ask for a summary before the chat fills up. The prompt in Keep a long conversation useful does that in one step. The new chat doesn't inherit the old one's history.

FAQ

Getting started

What does AI Chat do?

It's a chat for text. It answers questions, researches with web search, reads the files you attach, and writes or edits text. It has its own model picker and its own 5-hour limit. What is AI Chat? covers what it can reach.

How is AI Chat different from Super Agent?

AI Chat answers in the conversation. Super Agent takes on a multi-step job, uses tools such as web research, files and your connected apps to carry it out, and delivers the result, often as files. Use AI Chat to think the ask through and Super Agent to execute it. The two don't share a picker or a limit. Where AI Chat stops covers the handoff.

How do I choose a model?

Open the picker before your first message. On Plus and Pro from September 18, 2026, it has two groups, No credit cost and Flagship models. On every other plan it shows Recommended and then models grouped by maker. Each row shows what it costs. Which models cost credits, by plan explains the details. The controls are in Model picker.

What is Mixture-of-Agents?

It's listed under Genspark in the picker, and it runs several models on the same question before one aggregator writes the reply. See Mixture-of-Agents for when to use it.

Can AI Chat generate images?

No. It produces text only, though it reads images you attach. For a picture, use AI Image or Super Agent.

Can I use voice input in AI Chat?

On a computer, the mic button in the message box opens the Speakly site in a new tab. Speakly is Genspark's dictation app, and once it's installed it types what you say into any text box, including AI Chat. On a phone or tablet browser, the mic records your voice and turns it into text in the message box. Read the text before you send it, so a misheard word doesn't go through. Live voice conversation is a Super Agent feature.

Can AI Chat read my files?

Yes. Attach a PDF, Word, Excel, PowerPoint, text, code or image file and ask about it. The full list, the size limits and the Google export step are in Files and images.

Does AI Chat remember my past conversations?

No. It has no memory system, and it doesn't read your Hub files. Your chats are saved as projects, so ending a conversation doesn't delete it. Your Personalization profile is the one thing applied to every chat, and that's a profile you set, not a memory of what you discussed.

Plans and limits

Why did I hit a limit even though I have credits?

The 5-hour window and your credit balance are separate counters, and what happens when the window is spent depends on your plan. The four messages you can see, and what each means, are in AI Chat 5-hour limit.

Why were credits taken on a "No credit cost" model?

This applies to Plus and Pro plans from September 18, 2026, the plans whose picker shows "No credit cost". Three things cause it. The model you picked is a flagship model, which always bills credits. The pick is Mixture-of-Agents and the mix includes a flagship model, so the whole message bills. Or the 5-hour window was already spent, and core models were running on credits. The core section header then reads "Limit reached". On a free account every message uses credits.

How much does AI Chat cost?

It depends on the plan and on the model. Which models cost credits, by plan has one section for each case, including Team and Enterprise seats.

How do I turn off personalization?

Personalization applies what you've written in Settings → Personalization to every reply. Turn it off for a chat with the Personalization switch in the model picker, under Search Web. The switch appears once you've filled in your profile. Turn Personalization on or off has the steps.

What about my data?

The AI Data Retention switch is under Settings → Account. On a Team or Enterprise plan, the setting reads "Because you are on a Team or Enterprise plan, you are opted out of AI Data Retention."

Chats are kept until you delete them, and they don't expire. Deleting a chat removes it from your chat list, and it can't be opened again. To remove all of your data, use Delete User under Settings → Account. How and when deleted data is erased is described in the Privacy Policy.

Teams and admins

Does an Enterprise seat have the 5-hour limit?

No. Enterprise seats, and Team organizations billed by invoice, have no AI Chat limit. Team seats do have it, and they stop at it. The billing rule is in Team and Enterprise seats.

Why can't I delete a chat in my organization?

Your organization keeps project history, which turns deletion off. In the chat's "..." menu, Delete is replaced by "Your organization keeps project history. Deleting projects and conversations is disabled." An admin turns this on in Team admin › Policies › Project history › Keep project and conversation history. While it's on, files can still be edited or removed, and projects can still be removed from workspaces.

What does AI Chat cost on a Team or Enterprise seat?

The billing rule is in Team and Enterprise seats. Team & Enterprise Plans has seat pricing. Admins see credit usage in Team admin › Usage analytics, where Member Stats breaks usage down by member and by product, and Usage Logs can be exported as CSV. AI Chat messages that a seat covers use no credits, so AI Chat may show little or no credit usage there.

Can an admin control which models people use in AI Chat?

Yes, by vendor. You can't restrict individual models. In Team admin › Connectors, the AI Model Vendor Restrictions card has a Restrict AI Model Vendors switch. When it's on, members can only use models from the vendors you select, and AI Chat's picker hides the rest. If a chat was started on a model whose vendor is no longer allowed, the next reply comes from an allowed model instead. The card appears once vendor restrictions are enabled for your organization, so contact Genspark to turn them on.

Recipes

Thirteen recipes in five groups. The first six are the ones in the table at the top.

Choose the model for the job

1. Pick the right model for a job

For: a job you can describe in a sentence, where the answer comes back as text.

Prepare: the material you want it to work from, and the shape of the answer you want back.

Steps:

1. Open AI Chat and open the picker before you type.
2. Look at how the picker is laid out. If it opens on a No credit cost section followed by Flagship models, you're on Plus or Pro from September 18, 2026. If it opens on Recommended and then maker groups, that split doesn't apply to you, so skip to step 5.
3. Pick a core model for a first pass: summarizing, rewriting, explaining, drafting.
4. For a second opinion on the same job, pick a flagship model in a new chat. On the newer Plus and Pro plans flagship models use credits, and on some free accounts they show locked with an upgrade prompt.
5. Paste the prompt, then paste the contract text under it or attach the file.

Draft a one-page summary of this contract in plain language, then list the three clauses a non-lawyer should ask about.

You get: the answer in the chat.

If it stalls: if the answer misses the point, start a New Chat, pick another model, and paste the same prompt with one added constraint.

Related: Model picker, Which models cost credits, by plan, Use another model on the same topic.

2. Use Mixture-of-Agents on an unclear or high-stakes question

For: a question with no clean answer, where you want the trade-offs laid out before you decide.

Prepare: the question, written plainly, and the decision you're actually facing.

Steps:

1. Open the picker in a new chat.
2. Choose Mixture-of-Agents, listed under Genspark near the bottom of the picker.
3. Ask the question and say what balance you want.
4. Wait for the reply. It takes longer than a single model.
5. Check that the reply covers the objections, not only the case for one side.

Give me a balanced analysis of universal basic income. Present the strongest arguments on both sides, note where the evidence is strongest, and identify where the debate is most contested. Don't take a side.

You get: one reply written from several answers.

If it stalls: a slow reply is expected with a mix. For a fast answer, start a new chat and pick one model.

Related: Mixture-of-Agents, Model picker.

AI Chat MOA: one question answered by the mix, with both sides laid out

Ask about material and get text out

3. Ask questions about a PDF, Word or Excel file, or an image

For: a document you have to read and would rather interrogate than skim.

Prepare: the file, the question you need answered, and any number you want checked. A Google Doc, Sheet or Slide has to be downloaded first, as Word, PDF or Excel.

Steps:

1. Open a new chat and pick a model before you send anything.
2. Click the paperclip or + beside the message box and choose the file.
3. Name the file in the prompt, so the answer is tied to it and not to the chat in general.
4. Ask for the numbers explicitly when the answer depends on them.

Summarize the attached report in five bullets and list every number you used.

You get: the summary and a list of the numbers it used.

If it stalls: a scanned file comes back empty because AI Chat doesn't run OCR on attached files. Attach the pages as images to a model that reads images, or export a text-based PDF.

Related: Files and images, Research with web search and check the sources.

AI Chat Files: a PDF attached, read, and summarized with the numbers listed

4. Research with web search and check the sources

For: something that changed recently, where an answer from memory would be out of date.

Prepare: the question, the period you care about, and the kind of source you'd trust.

Steps:

1. Open the picker and switch Search Web on before the first message.
2. Ask the question with the period in it. "The past 6 months" is read differently from "recently".
3. Read the [Source] link beside each claim. Search Web adds them on its own.
4. Open two or three of the links and check that each source says what the reply says.
5. Leave search on for the rest of the chat.

Search the web for the latest research on longevity and anti-aging science and summarize the key findings from the past 6 months. Include links to your sources.

You get: an answer with a source for each claim.

If it stalls: if the Search Web switch is grayed out, that model can't search. Start a new chat with another model.

Related: Web search and sources.

5. Draft or rewrite text

For: an email, a post or a paragraph that needs a different tone, a shorter length or a cleaner structure.

Prepare: the text, the reader it's for, and the one thing it has to say.

Steps:

1. Paste the text into the message box and put the instruction above it.
2. Name the tone and the limit, because "friendlier" alone gives you a longer reply.
3. Send it and read the result against your own voice.
4. To change one part, use the pencil on your message, edit it and send again. The reply is regenerated from there.
5. Use Copy under the reply to take it out.

Rewrite this email to sound friendlier and keep it under 120 words.

You get: a rewrite in the chat, with your original text still in your message above it, unless you edited that message with the pencil.

If it stalls: a rewrite that drifts from your meaning means the prompt didn't name the thing to keep. Say "keep the request and the deadline, change the tone".

Related: Personalization, Save an answer.

6. Shape a prompt, then hand it to Super Agent

For: you know the outcome you want and you can't write the brief that gets there.

Prepare: the idea, roughly, and the constraints you already know.

Steps:

1. Describe the idea in a couple of sentences, in a new chat.
2. Paste the prompt below after your idea. It asks you what is missing before it writes anything.
3. Answer the questions it asks. Those answers are the brief.
4. Ask for the final prompt as one block you can copy.
5. Use Copy, open Super Agent, and paste it. This step is manual.

Help me turn this idea into a clear brief for an agent: ask me what is missing, then write the final prompt.

You get: a prompt with the gaps filled in, ready to paste into an agent that can run it.

If it stalls: if it writes the prompt without asking anything, tell it which parts are still open, such as the audience or the output format.

Related: Where AI Chat stops.

AI Chat Shape: an idea turned into a brief, copied across to Super Agent as a prompt

When the model changes

7. When a model is gone

For: a model you used is no longer in the picker, or a chat shows a notice that its model will be retired.

Prepare: nothing, though the chat where you used it helps.

Steps:

1. Read the notice above the message box: "{model} will be retired on {date}. Switch to {replacement} to avoid interruption." It appears only in chats that use the retiring model. You can dismiss it, and it comes back in the model's last 7 days. Troubleshooting explains why models retire.
2. Start a New Chat and pick the replacement, or keep going in the old chat. After the retirement date, its next reply comes from the replacement automatically. Either way, messages are billed at the replacement's rate, which may differ from what the old model cost.
3. Keep the old chat for reference. Its history stays in your chat list.
4. If the replacement behaves differently, say what changed in the prompt, such as the length or the format you got used to.

You get: a working chat on the replacement model, and the old conversation still readable and, after the retirement date, able to continue on the replacement.

If it stalls: if the replacement isn't in your picker, it may not be offered on your plan, in your region, or by your organization's model policy. Pick another model in the same group, or ask your admin. Retired models in Troubleshooting has the background.

Related: Use another model on the same topic.

8. Use another model on the same topic

For: the answer was thin, wrong or badly written, and you want a second opinion.

Prepare: the prompt that got the weak answer.

Steps:

1. Copy your original prompt from the old chat.
2. Open New Chat. The picker in the finished chat is grayed out, so a new chat is the only way to change the model.
3. Pick a different model, in the other group if your picker shows groups.
4. Paste the prompt and add what the first answer got wrong.
5. Read both answers. There's no side-by-side view, so the comparison is yours to make.

You get: a second answer, and the first still in its own chat for reference.

If it stalls: if both models give the same weak answer, add the constraint you left out and ask again.

Related: Model picker, Pick the right model for a job.

Keep a chat useful and find it again

9. Keep a long conversation useful

For: a chat that has run for hours and has started to lose the early detail.

Prepare: the decision or the output you need to carry forward.

Steps:

1. Watch for "For better performance, previous chat history has been compacted." That's the chat making room, and it's normal.
2. Ask for a summary before you start fresh, while the chat still has the thread.
3. Start a New Chat with the summary as the first message.
4. Keep one topic per chat. A new topic gets a new chat from the start.

Summarize what we decided so far in 10 bullets so I can start a new chat.

You get: a carry-over summary you can paste into the next chat, without rereading the old one.

If it stalls: if the reply stops with "Context Length Exceeded", the chat has nothing left to work with. Start a new one, even without the summary.

Related: Long chats, Save an answer.

10. Save an answer

For: a reply you'll want next week, outside the chat.

Prepare: nothing beyond the reply.

Steps:

1. Use Copy under the reply for anything you'll paste elsewhere.
2. Open the "..." menu under the reply for the two save options.
3. Choose Save to AI Drive to keep the reply as a Markdown file in the Genspark › AI Chat folder of your AI Drive. On a free plan, this may ask you to upgrade first.
4. Choose Save to Notion to put it in a page. Until Notion is connected, that row reads Connect Notion and starts the connection first.

You get: the reply saved to AI Drive or Notion, and the chat unchanged. A saved reply is named after the chat, with the date and time added.

If it stalls: after a save, View in AI Drive on the Export Successful dialog opens the file. If Save to AI Drive asks you to upgrade, your plan doesn't include saving files out of a chat, so use Copy instead. If the Notion row reads Connect Notion, connect Notion first and save again.

Related: Find, continue, pin, rename or delete an earlier chat.

11. Find, continue, pin, rename or delete an earlier chat

For: a conversation you had a while ago and need again.

Prepare: a word from the title, or roughly when you had it.

Steps:

1. Open the chat list in the sidebar. Every AI Chat conversation is saved there as a project.
2. Open the chat to continue it. If a saved chat opens in a viewer, the button reads Continue Conversation.
3. In the chat list, choose Pin for a chat you return to, so it stays at the top. The limit is 50: "You can only pin up to 50 projects. Please unpin some first."
4. Use Rename to make a chat findable, Bookmark to mark it, or Delete to remove it. Delete appears only on chats you own.
5. Add to Hub adds the chat to your Hub.

You get: the chat open and ready to continue on its original model (or on its replacement, if that model has been retired), or a cleaner list.

If it stalls: if you can't delete a chat, your organization may keep project history. The chat's menu then shows "Your organization keeps project history. Deleting projects and conversations is disabled." where Delete would be.

Related: Long chats.

Dictation and personalization

12. Dictate on the web

For: you'd rather speak the prompt than type it.

Prepare: on a computer, the Speakly app. On a phone or tablet, a browser that allows microphone access.

Steps:

1. Open a chat on the web and click the mic in the message box. On a computer, this opens the Speakly site in a new tab. Install Speakly from there, then come back to the chat. On a phone or tablet, recording starts in the message box.
2. Dictate your prompt. On a computer, Speakly types it into the message box. On a phone or tablet, the voice input turns it into text when you finish.
3. Read the text before you send it, and fix any misheard word.
4. Send it as you would a typed message.

You get: the prompt as text, without typing it.

If it stalls: on a computer the mic always opens the Speakly site and doesn't record in the page, because desktop dictation runs through the Speakly app. Speakly covers how it works. Live voice conversation is a Super Agent feature.

Related: Speakly, Realtime Voice.

13. Turn Personalization on or off

For: replies that should match your tone, or should stop matching it.

Prepare: a profile, if you want the switch to appear.

Steps:

1. Open the picker dropdown in a chat.
2. Find Personalization under Search Web, labeled "Apply your profile preferences".
3. Switch it off for this chat.
4. To change what it applies, open Settings → Personalization and edit the profile.
5. To check the effect, ask the same question in two chats with the switch in each position.

You get: replies shaped by your profile, or replies without it, chat by chat.

If it stalls: if the switch isn't there, there's no profile yet. Fill one in under Settings → Personalization first.