How to turn a ChatGPT screenshot into a weekly check
An image in Slack is one session. Recover the question, freeze a small list, and let the next runs tell you whether the answer repeats.
Built by StyleLine FZE in Sharjah. Email support@mentionclock.com. We reply within one business day.
Choosing a tool is a different job from a single method. The dated vendor scorecard is best ChatGPT brand monitoring tools.
Sales will forward a picture of a chat where a rival is named and you are not. Sometimes the picture is flattering and someone wants it in a deck. Either way, a ChatGPT screenshot is a single session with a date you may not even have. A weekly check is the same question, asked again, with the brands and the links written down. The job is to convert the image into that question without pretending the image was already a trend.
Mention Clock is built for the repeat, not for reading pixels in a PNG. You still start from the image, because that is the artifact people trust. Then you move the wording onto a project. Paid plans query ChatGPT, the second chat engine, and Perplexity through official APIs. Free covers two of those engines, on 5 prompts, weekly, at $0, with no card. The upcoming engine is being added. A picture of the upcoming engine, or of a Google AI Overview, can sit in the same inbox and must not be filed as a ChatGPT result. Mention Clock does not scrape Overviews.
Read the image before you react
Open the file at full size. You are looking for five facts, and you should write "not in the image" when one is missing.
Product. ChatGPT, the second chat engine, Perplexity, the upcoming engine, or a Search overview are different products. The logo, the URL bar, or the citation style usually gives it away. Perplexity shows numbered sources. An Overview sits on a results page. If you cannot tell, ask the sender which site they were on. A wrong product name poisons every later comparison.
Prompt. The buyer's question must be visible, including constraints such as team size or industry. A cropped image that shows only the answer is not enough. Reply and ask for the prompt before you brief anyone. If the sender retyped it from memory, label your note "reconstructed" so nobody treats it as verbatim.
Date. Use a timestamp in the product if you can see one. If the only date is when the file hit Slack, write "received [date]" and do not call that the answer date. Chats can be old.
Names in order. List every brand in the answer, in the order they appear, including yours if it appears. "We were missing" is incomplete if you do not record who was present.
Links. Copy any URLs or hosts that are visible. Many screenshots cut the sources off. Ask for a second image of the citations when the first one is only the paragraph. A name without a source is a different problem from a source that recommends someone else.
Also note whether the session looks logged in. A personal account with memory is not a clean test. That does not make the image useless. It means you should not be surprised when a fresh run disagrees.
Decide what the image is allowed to prove
The image proves that one session produced that answer under conditions you only partly know. It does not prove that "ChatGPT recommends Rival" as a standing fact. It does not prove the opposite when the image is friendly. Put the image in a folder with the five facts beside it. Do not paste it into a board slide titled with a rank.
If the sender is upset, acknowledge the prompt and the rival names. Then say the next step is to ask that prompt again on a schedule. People calm down when the follow up has a date. They do not calm down when you argue with a screenshot using another screenshot from your own phone. Your phone is a second anecdote.
API answers and consumer apps can differ because apps personalize. Mention Clock's chart is a repeated survey. When the weekly rows and the original image disagree, keep both, and label the image as the session that started the check. Do not average a phone and an API into one cell.
Build the list the image was hinting at
Take the recovered prompt as prompt one, verbatim. Then add four siblings from real conversations, not from a keyword tool's idea of your category. A practical set:
- The screenshot prompt, unchanged.
- The category question buyers ask when they do not know your name.
- The problem question, in the words a customer used on a call.
- An alternatives question that names the rival in the image.
- A comparison that names you and that rival, if both names belong in a public answer.
Five is the free project cap, so this set can run without a paid plan. If you already have a longer list on Starter (25 prompts, $29 a month or $290 a year) or Pro (100 prompts, $79 a month or $790 a year), add the screenshot prompt to that list instead of starting a second project. One brand, one project. Agency is $199 a month or $1,990 a year when you need up to 15 projects. Yearly billing is 10 months. Details stay on pricing.
Do not add five paraphrases of the screenshot prompt. Paraphrases look like diligence and break the comparison, because you will not know which wording the buyer used. One verbatim line plus four distinct jobs is the set.
Freeze the text. If product marketing wants a "better" prompt that mentions your differentiator, save that as a draft page on your site, not as a replacement for the buyer's wording. The check is supposed to imitate the buyer who took the screenshot.
Run it in a way you can repeat
Clean pass, once, today. If you are doing it by hand, use a test account without custom instructions. Paste prompt one. Record named or not, order, rivals, and hosts. Repeat for the other four. The columns match the tracker:
Set the engine cell to the product you actually opened. If the screenshot was ChatGPT, the hand pass is ChatGPT. Do not "also try the upcoming engine while you are here" and mix the rows. The upcoming engine hand pass can live in its own rows. The product status for that engine is covered on The upcoming engine brand mentions: the upcoming engine is being added, and a hand log is the honest method until a scheduled run includes it.
Scheduled pass, every week. Put the same five strings on a Mention Clock project. Pick engines on purpose. If the image was ChatGPT, include ChatGPT. Use the second free engine for the assistant your buyers also open, often the second chat engine or Perplexity, and keep those rows separate. Starter includes all three shipped engines and Slack or Teams. Email exists on every plan, including a note when you disappear from a prompt you had been winning.
Wait for four weekly runs before you describe a pattern. The original image can sit in an appendix of the first weekly AI visibility report. It should not be the only evidence in week four. By then you have a history. Show the history.
How to brief the person who sent the picture
Send a short reply the same day so the screenshot does not become folklore.
Say which product you think it is, and ask them to correct you. Paste the prompt you recovered. List the brands you saw, in order. State the date you are using. Tell them the prompt is now on a weekly run, and name the engines. Offer the day you will send the first comparison. Do not promise that the next answer will include your brand. Promise that you will measure this wording and look at the pages those answers cite.
If they ask "what do we do about it" before any second run, give them the cited host from the image if you have one, or say you are waiting for a full answer with sources. Inventing a content task from a cropped paragraph creates busywork. A host you can open is a task. A vibe is not.
When the first scheduled run finishes, reply again with three lines: named or not on the screenshot prompt, who else was named, and the host. Link the longer note if they want it. That second reply is what teaches the team to send you prompts instead of unmarked images.
Edge cases that show up constantly
The image is Perplexity. Keep the prompt, and judge the numbered list with how to check if Perplexity cites your brand. A Perplexity paragraph and a ChatGPT paragraph are not interchangeable. Run both only if both products are in the plan, and report them as two columns.
The image is an Overview. Do not import it into a ChatGPT series. Note the query, the date, and whether you were named, beside the chat runs, the way AI Overviews brand mentions describes. The chat project still has value, because the pages cited in chat are often the pages worth fixing. It is not an Overview rank.
The image names a feature you retired. That is an accuracy issue, which Mention Clock flags as a stale price or a dead feature when the run sees it. On day one, write the wrong claim next to the prompt and point at the URL that still says the old thing. Fix that URL even if you never run a second prompt. The screenshot did its job.
Several people send conflicting images in one week. File each with its prompt and product. Do not pick the scary one as the truth. If the prompts differ, you have several buyer questions, which is useful. If the prompt is identical and the answers differ, you have variance, which is why a weekly repeat exists. Four runs will not be identical either. Look for whether you are usually named, not for a single verdict.
The sender wants it in the board appendix tomorrow. Give them the five facts and label the slide "one session, received [date], prompt recovered." Add a line that a weekly repeat has started. Refusing the appendix creates a shadow copy you do not control. Labeling it keeps the anecdote in its lane.
What you store after the image is filed
On each later run, keep mention, place, rivals, and cited pages. Share of mentions belongs in the note only when the rival set is explicit. The formula and a real run dated 28 Oct 2026 are on share of voice in AI search. Tone and wrong facts belong in the note when they would mislead a buyer. Alerts worth opening are the ones the product already sends: you vanished from a prompt you used to win, a rival showed up for the first time, share moved under a target you set, or a new source about you was cited. You do not need a new inbox rule for screenshots once the prompt is on the project. Ask people to send the prompt text. Accept images anyway, because that is how the mail actually arrives.
The hand method for a full ChatGPT list, beyond the five lines a screenshot inspires, is how to track brand mentions in AI chats. Use it when the image convinced the team that the category is worth more than one question. Until then, five honest prompts beat a library nobody reruns.
Start without a meeting
Create the project with the screenshot prompt and the four siblings when you are ready for the clock to take over: create a free project. If you want three sample questions on one engine before you commit the wording, use the AI visibility checker and then replace those samples with the buyer's real prompt. The checker is a preview. The screenshot's wording is the asset worth freezing.
Related guides
- How to track brand mentions in AI chats for a longer ChatGPT list
- How to write a weekly AI visibility report for the mail a boss can forward
- How to check if Perplexity cites your brand when the image was Perplexity
- Resources for the full guide set
FAQ
Is a screenshot enough to report a rank?
No. It is one session. Recover the prompt, record the brands you can see, and run that wording again on a schedule before you describe a pattern.
What if the prompt is cropped out?
Ask the sender for the question they typed. If they only remember a paraphrase, mark your note as reconstructed and do not treat it as the original wording.
Can I track the screenshot prompt on the free plan?
Yes. Free is 1 project, 5 prompts, and 2 engines, weekly, at $0, with no card. Put the recovered prompt in that set. Paid plans add the third engine and a higher prompt cap. The upcoming engine is being added and is not a scheduled engine yet.
What if the image is the upcoming engine or an AI Overview?
Label it as that product. Do not file it as ChatGPT. The upcoming engine can be logged by hand until Mention Clock includes it. Overviews are a manual Search note. The scheduled engines today are ChatGPT, the second chat engine, and Perplexity, with two of those on Free.