Frequently asked questions

• More people now ask AI assistants like ChatGPT, Claude, and Perplexity for recommendations: “What's the best accounting software for a small agency?”
• PromptPulse asks those assistants the questions your customers ask, on a schedule, and records whether they recommend you, how highly, in what tone, and which websites they rely on.
• It does the same for the competitors you choose, so you can see who is winning the answers.
• Then it shows you where competitors get recommended and you don't, and what to do about it: the Fix-It Playbook.

• An AI answer usually names only a handful of businesses. If you're not one of them, that buyer may never see you at all.
• You can't see these answers from the outside. Each person asks privately, and the answer can differ each time. Measuring is the only way to know what buyers are being told.

• Related, not the same. Search engine optimization (SEO) is about ranking in a list of links. Here, the AI reads pages for the buyer and writes one answer.
• Good SEO helps, because AI engines search the web too. But they weigh what other sites say about you heavily — reviews, comparisons, forum threads — not just your own pages.
• This work is often called generative engine optimization, or GEO.

• A prompt is a question typed into an AI assistant — here, one a potential customer would ask.
• Track questions buyers ask before they know your name: “best payroll tool for a 20-person company”, not “is [your brand] good?”. Those are the answers you need to win.
• Setup suggests starter prompts for your industry. Edit, remove, or add your own at any time.
• A handful of prompts that match how real customers talk beats a long list of near-duplicates.

• Add the businesses a customer would actually weigh against you for the same need — not every brand that shares a word or a category with yours.
• Your Visibility Score doesn't change with your competitor list. It measures only how answers treat you, and competitors are never added to the questions the engines are asked.
• Share of voice does. It's your share of the answers that name you or any competitor you track. A household name that isn't a real rival, but that answers mention often, pulls your share down.
• So does the Fix-It Playbook. It lists sites cited when a competitor was recommended and you weren't, so a wrong competitor can surface sites that don't matter to you.
• Edit competitors at any time; the next run measures the new list. A competitor added after a run shows as “Not measured” for that run, and removing one removes its results too.

• There's no universal good number. Scores depend on your industry, how crowded it is, and which prompts you track.
• 100 would mean every answer names you first, speaks positively, and links to your site. Very few brands come close on broad questions.
• Two comparisons matter more than the number itself: you versus your competitors (the competitor table) and you versus your own past runs (the trend chart).

• Usually not. Many businesses, especially smaller or newer ones, aren't named by AI engines yet. That's the gap PromptPulse is built to show and help close.
• Check that your brand name and website on the project match how customers and other sites write them.
• A score like 4 ± 4 means too few mentions to tell apart from none — see What the ± margin means.
• If an engine failed during a run, it's left out and the run is marked incomplete. A failure never shows as a zero.

• No. Setup is your business name, website, competitors, and the questions customers ask.
• Most suggested actions happen off your site: claiming a review profile, asking a writer to add you to a list, joining a community thread. No code changes needed.

A run is the same five steps every time:

• Ask. Each prompt goes to each engine through the provider's official API, as a brand-new user would ask it: no chat history, no memory, no account, and no extra instructions from us. Web search is on, as it is in the consumer apps.
• Repeat. Each prompt is asked three times per engine by default (adjustable from 2 to 5 per prompt), because the same question can get different answers.
• Read. A separate analysis model reads every answer and records which tracked brands it names, in what order, in what tone, the reasons it gives, and the sources it links to.
• Check. Every linked page is visited to confirm it exists. Dead or made-up links are left out.
• Score. Scores, share of voice, and citation gaps are calculated from those records.

No scraping or browser automation is used at any step.

• Each model is the one a regular user of that app gets by default — never a cheaper model swapped in to save cost, which would measure something your customers never see.
• The analysis model that reads the answers is claude-haiku-4-5-20251001. It only reads and records; it's never one of the engines being measured.
• Providers update their models over time. Every answer is stored with the exact model version that produced it, listed under each run's engine cards.

• They cover both kinds of AI assistant. ChatGPT and Claude are general-purpose assistants people ask about anything, including what to buy. Perplexity is built around search: every answer is a summary of pages it just looked up.
• Perplexity always shows its sources. Every answer links the pages it drew on, which gives the Fix-It Playbook the most to work with.
• Gemini is off by default, not left out. Every engine on a prompt is more answers to collect on every run, so it's on where your customers use Google's assistant, not everywhere by default.

• Engines are chosen per prompt, on the prompt form. Every prompt needs at least one.
• Your plan price is the same whichever engines you choose.
• A change applies from the next run. Past runs keep what they measured.
• Your overall score is the average of the engines measured, so adding or removing one moves it. Compare engine by engine to see real change.
• A run finishes when its slowest engine does, so adding Gemini can make runs take longer — see why some engines take longer.

• Perplexity is asked directly, one question at a time, so its answers usually arrive within minutes.
• OpenAI, Anthropic, and Gemini are sent through each provider's batch service: the whole set of questions is handed over at once, and the provider answers it when it has spare capacity. That can take minutes or many hours, and providers allow up to 24.
• Batch processing costs about half as much as asking directly. That saving is a large part of what keeps PromptPulse's pricing where it is.
• The wait depends on the provider's workload that day, not on your brand. A slow engine says nothing about your results.
• A run finishes when the slowest engine's answers are back and analyzed. The progress view shows each engine's status while you wait. After Run now, it's safe to close the page: you'll get an email when the run finishes.

• Your own account remembers you: past chats, saved memory, your location, sometimes your own business. The answer you see is shaped for you. PromptPulse asks as a new user with none of that.
• The same question gets different answers from one try to the next. One check is one draw.
• Paid plans in the consumer apps can use a different model than the free default measured here.
• That's why PromptPulse asks every question several times and reports an average with a margin of error.

• Normal variation. Answers vary run to run. A change inside the ± margin isn't a real change.
• The web changed. A new review, article, or forum thread — about you or a competitor — can shift what engines find.
• The model changed. Providers update their models. Check the model versions listed under the engine cards.
• Your setup changed. New prompts or competitors change what's measured. Scores are only compared between runs that measured the same prompts.
• Compare runs on your project page shows which changes are real.

score = 100 × ( presence × (0.4 + 0.35 × rank + 0.15 × tone) + 0.1 × links )

Each has its own article with worked examples: Visibility Score, margin of error, share of voice and reasons, and the Fix-It Playbook.

No provider publishes exactly how its engine chooses. What can be said, and what PromptPulse can observe:

• What it already knows. Each model learned from a large snapshot of the web before its release. Well-known brands start ahead. This only changes when a new model comes out.
• What it finds right now. With web search on, the engine looks up pages for each question and builds its answer from them. These are the sources it links to — and the part you can influence soonest.
• Who else vouches for you. The sources engines cite are mostly independent — reviews, comparisons, forum threads, press — rather than a business's own site.
• How well you fit the question. “For a 10-person team” or “under $50” changes the answer. Clear, findable facts about who you serve, what you cost, and what you do help an engine match you.

The reasons answers actually give, which PromptPulse counts for you and each competitor:

• Pricing or value
• Ease of use
• Features
• Integrations
• Fit for the asker's size, industry, or use case
• Reviews and reputation
• Support and onboarding
• Scalability and performance
• Security and compliance
• Product quality

These are the kinds of sites AI engines cite most when recommending businesses:

Which sites matter differs by industry and by engine. The Fix-It Playbook lists the ones actually cited for your prompts.

• They're independent. A review or forum thread is someone other than the business vouching for it — a second opinion an answer can point to, not the company's own claims.
• They reflect real experience. Reviews and community threads describe what using a product is actually like, including the downsides.
• They already compare. A “best of” list or comparison page answers the exact question the buyer asked, ready to summarize.
• They're easy to find and stay current. These sites rank well in web search and are updated constantly, so they show up when an engine searches.
• Your own website still matters: links to it count toward your score, and it's where engines check facts about you.

• If you don't sell software, G2 isn't for you. The review sites that matter are the ones your customers already use, like Yelp or a marketplace's buyer reviews.
• Reddit fits almost any business, through the communities your customers belong to: your city's subreddit, a parenting forum, a hobby group.

• Each engine runs its own web search, so one may cite Reddit for a question where another cites a news article instead.
• Which sources engines favor changes over time. Weekly runs show you when it shifts.

• The Fix-It Playbook lists only gaps: sites cited when a competitor was recommended, but never when you were. A Reddit thread cited when you were recommended too isn't a gap, and that's good news.
• The Reddit and review sites panel on your project page shows whether those sites were cited at all in your latest run, and whether those answers recommended you, a competitor, or no brand you track.
• If they weren't cited at all, engines found other sources more useful for your questions. That isn't a problem. Spend your time where the playbook points.
• A site that blocks automated link checks, as Reddit often does, still counts. Only links to pages that are gone or broken are left out.

• On Reddit, take part as yourself. Answer questions you can genuinely help with and say who you are. Undisclosed promotion breaks most communities' rules, gets removed, and can backfire.
• For reviews, ask real customers. Never buy or write fake reviews. It breaks the review sites' rules and the FTC's.
• Either way, it takes weeks for engines to pick it up — see how long changes take.

• Usually weeks. Engines pick up new pages and reviews over days to weeks as they search the web.
• What a model already knows only changes when the provider releases a new model, often months apart.
• Watch the next two to four weekly runs rather than re-running right away — see Getting the most from your runs.

• Not through PromptPulse: we don't sell placement in AI answers. What we measure is the engines' ordinary answers to your customers' questions.
• What works is earning mentions on the sites engines read — real reviews, accurate listings, being included in comparisons — which is what the Fix-It Playbook points you to.
• Avoid fake reviews and undisclosed self-promotion. Sites remove them, and they can hurt your reputation where engines look.

• The AI answers at the top of Google Search have no official API. Measuring them would mean scraping Google, which PromptPulse doesn't do.
• Gemini, Google's AI assistant, can be measured: turn it on for any prompt where your customers use it.

• Engines receive your prompt text only — not your name, email address, or billing details.
• The analysis model also receives the brand and competitor names you track, so it can find them in answers.
• Setup suggestions (starter prompts, competitors) use your website and business description.
• We don't sell your data or use it to train AI models.
• Full answer text is deleted after 90 days. The scores and findings drawn from it are kept, so your history stays intact.
• Details are in the Privacy Policy.

Email support@promptpulse.io.