How do I check my brand's sentiment across AI engines?

Q: Which prompts surface AI sentiment about a brand?

Use a description prompt like 'What is [brand] known for, and what are its strengths and weaknesses?' and a comparison prompt like 'Compare [brand] to [competitor 1] and [competitor 2] and recommend one.' The comparison prompt usually exposes harsher relative sentiment than the standalone description.

Q: How do I read tone in an AI answer?

Look at word choice ('trusted' vs 'lesser-known'), the number of caveats and hedges, what the engine praises versus what it goes quiet on, and where your brand appears in the answer. These cues reveal sentiment even when the surface recommendation looks neutral.

Q: Why does AI sentiment about my brand keep changing?

AI engines sample probabilistically, so the same prompt can return a warmer or cooler answer run to run, and their training and retrieval data keep shifting. Treat a single reading as a snapshot, run each prompt a few times to see the spread, and track the trend over weeks or months rather than reacting to one wobble.

Question

How do I check my brand's sentiment across AI engines?

Maciej Grabek · Accepted Answer

Ask each engine to describe your brand and to compare you to competitors, then read the tone of the answer: what it praises outright, what it hedges, and what it warns against. Log each result as positive, neutral, or negative per engine. Do this on the same day across ChatGPT, Perplexity, Gemini, Claude, and DeepSeek, and repeat it over time because sentiment shifts run to run. Run the same two prompts in every engine Sentiment lives in how an engine talks about you, not just whether it names you. To surface it, you need prompts that force the model to characterize your brand rather than list it. Open a fresh chat (no prior context) in each engine and ask the same two questions. A description prompt pulls the model's unprompted framing: "What is Acme Payroll known for, and what are its strengths and weaknesses?" A comparison prompt exposes relative sentiment, which is often harsher than the standalone view: "Compare Acme Payroll to Gusto and Rippling for a 20-person startup. Which would you recommend and why?" Run both in all five engines: ChatGPT, Perplexity, Gemini, Claude, and DeepSeek. Use a fresh session each time so one answer does not prime the next. Read the tone, not just the verdict Once you have ten answers, ignore the surface recommendation for a moment and read how each engine words things. The signal is in the qualifiers. Word choice: "trusted" and "well-regarded" are positive; "lesser-known" and "newer entrant" are neutral-to-negative framing even when the facts are flattering. Caveats: count the hedges. "Good, but their support has mixed reviews" is a different sentiment than a clean recommendation. What it praises vs what it hedges: an engine that praises your pricing but goes quiet on your product depth is telling you where its training data is thin or skeptical. Order and prominence: being named third after two competitors, with a shorter description, is a weaker signal than leading the answer. Log positive, neutral, or negative per engine Score each engine on a simple three-way scale and keep the receipts. A basic grid works: engine down the side, the two prompts across the top, a rating and a one-line quote in each cell. Paste the exact answer text into a note so you can compare wording later. Rate each answer positive, neutral, or negative based on the tone cues above. Flag any hard factual errors separately - a wrong founding date or a confused product category is a fixable content problem, not sentiment. Note which sources the engine cited, especially in Perplexity, which shows its links. Negative sentiment usually traces back to a specific review site or thread. Read a real example of phrasing revealing sentiment Sentiment is easiest to see when two engines describe the same brand differently. Suppose you ask both Claude and DeepSeek about a mid-market CRM: Claude: "It is a solid, established option for teams that want deep customization, though users often mention a steep learning curve." DeepSeek: "It is a powerful but complex platform that many small teams find overwhelming and expensive." Both name the same trait - complexity. Claude frames it as "deep customization" with a mild caveat, landing at neutral-positive. DeepSeek frames it as "overwhelming and expensive," landing at negative. Same fact, opposite sentiment. That gap is your work: the negative framing usually comes from the sources that engine leans on, so the fix is off-site (better reviews, clearer positioning content), not a prompt tweak. Track it over time, because sentiment is fuzzy Sentiment is the least stable signal in AI search. The same prompt can return a warmer or cooler answer from one run to the next, because these models sample probabilistically and their training and retrieval data keep changing. A single reading is a snapshot, not a verdict. Run each prompt two or three times per engine to see the spread, and re-check on a fixed cadence - monthly is enough for most brands, weekly if you are actively shipping content to move it. What matters is the trend: is the neutral engine warming up, is the negative one still anchored to one bad thread. Chasing a single-run wobble wastes effort. Doing this by hand across five engines, two prompts, and a few repeats is a couple of hours a month, and the scoring is subjective. If you want it standardized and charted over time, avisibli runs your prompts across all five engines on a schedule and scores the sentiment of each answer so you can watch the trend instead of re-reading transcripts. avisibli is the GEO platform that publishes this answer library. Self-references are limited to topics where a tool-based answer is genuinely useful to readers.