Frequently Asked Questions

Start with three things:

1. Review your searches — these are the phrases indextr uses to test your AI visibility. Make sure they reflect how your customers actually describe what they need. Add or adjust them under the Searches tab.
2. Add competitors — add your main competitors under the Competitors tab so you can see how often you are named alongside the businesses you actually compete with.
3. Check your recommendations — once your first scan completes, go to your dashboard and start with the top recommendation on your list. Each one shows when it is worth checking again.

AI visibility measures how often and how prominently AI engines — ChatGPT, Google AI Overview, Gemini, Perplexity, and Claude — recommend your business when someone asks a relevant question.

indextr runs a set of search phrases through the AI engines your plan covers and records whether your business was mentioned, whether your website was cited, how you were described, and how you compared to competitors. After each scan, your Visibility Score is updated and new recommendations are generated based on what the scan found. Your dashboard is updated after each scan.

Searches are the phrases someone might type into an AI engine when looking for a business like yours — for example, “best estate planning attorney in Denver” or “affordable HR software for small teams.” Choose phrases that reflect how your actual customers describe what they need, not how you describe what you sell. indextr can suggest phrases based on your business description, or you can add your own.

Competitors give indextr a reference point. During each scan, it checks whether your competitors appear in the same AI responses as your tracked searches. This shows you not just whether you showed up, but how often you appeared alongside the businesses you actually compete with. It does not affect your Visibility Score.

Your score is a weighted average across the AI engines your plan covers — ChatGPT on Free; ChatGPT, Gemini and Google AI Overview on Starter; all five including Perplexity and Claude on Pro and Done-for-You. Each engine is weighted by its estimated share of consumer AI search, with ChatGPT and Google AI carrying the most weight and Claude the least. Within each engine, your score is based on how often you were mentioned and how high up you appeared, whether your URL was cited, and how positively you were described. Being named and how prominently are counted as one measure, and it is the largest part of the score. How you compare against tracked competitors is measured and shown, but it does not affect the score.

They measure different things. Mentioned In is a straight count — the share of all the answers we collected where your business was named, with every result counting equally. Your Visibility Score weights each engine by our estimate of how much it is used, so being named only by a smaller engine lifts your mention rate without moving your score much. The score also folds in how high up you appeared, whether your site was cited, and how you were described. A high mention rate with a low score can mean you are visible mainly on the engines with the least weight.

Each scan runs every phrase you track through every AI engine on your plan, so three phrases on Starter is nine answers. Sometimes fewer are counted, and you will see something like “Across 8 of 9 answers”. One reason is that Google does not show an AI Overview for every search, so for that search there is no answer to read. Rather than record it as “you were not mentioned”, indextr leaves it out of the count: an answer that never appeared is not evidence you are invisible. The same applies when an engine replies by asking where you are instead of answering. Counting those as misses would report a worse score than the truth.

AI engine responses vary naturally from run to run — the same question can return different answers on different days as models update and new content gets indexed. A single-scan change of a few points is normal. A consistent trend in one direction over multiple scans is the meaningful signal to watch.

AI search results can be personalized — based on your signed-in Google account, search history and location — and the same person searching the exact same phrase twice can get different results.

indextr scans from a consistent baseline — no signed-in account and no search history — so every scan is measured the same way every time. That makes it possible to track a consistent, comparable signal over time, rather than a snapshot of any one person's session.

If you want to sanity-check a result yourself, search in Guest Mode (or a signed-out incognito window) rather than your regular signed-in browser. That removes your own personalization and should align much more closely with what indextr shows.

What indextr is built to show you is the trend — whether your visibility is going up or down across many scans and search phrases, and how you compare to competitors — not a promise that any single scan will exactly match any single search you happen to run yourself.

A score of 0 means your business was not mentioned in any of the AI responses for your tracked searches. Your recommendations are specifically designed to move this number.

Your search phrase exactly as you typed it — the same thing a person types into the box — plus your business's city as the searcher's location when the phrase says “near me” (see the next question). We deliberately don't attach hidden instructions telling the AI how to answer. Sending the phrase as written means what you see in indextr is something you can go and check yourself.

Because the phrase decides what you're measuring, and that's your call rather than ours to guess. There are three cases:

• You name a city — “smart home installer denver co” — that's a local search, and we send it exactly as written.
• You say “near me” — that's a local search too, but without a place in it. A person typing that into ChatGPT has their location filled in automatically from their internet connection; an API request has none, so we supply your business's city the same way. Without it, ChatGPT would just ask which city you're in and name nobody.
• You name no location at all — “what does smart home installation cost” — we send nothing extra, because that's a national question and adding a city would change what you asked.

We never rewrite your phrase in any of these cases. The words we send are the words you typed.

ChatGPT can give a different answer each time it is asked, so your check and our scan are two answers from the same engine rather than one of them being wrong.

For some phrases, your position genuinely moves. On a competitive search, ChatGPT might put you first today and fourth tomorrow. There simply isn't one fixed answer to find. Check it twice yourself and you can disagree with your own first result.

For those phrases, visibility is better understood as how often than as a single position. If ChatGPT names you in two of every three answers, then one time in three someone asking won't see you — so a check that comes back empty isn't a contradiction, it's that one in three. It's why the Last 3 Scans column shows “2 of 3” rather than a single verdict.

Two things can widen a gap. If you're signed in, ChatGPT can remember what you've discussed before, so your chat history can shape its answer. And the ChatGPT app isn't the same service indextr queries, so some difference is built in.

To compare fairly: open a Chrome guest window, stay signed out of ChatGPT, and use a fresh window each time. If the phrase is one where your position moves, check three times and compare the pattern against your Last 3 Scans rather than one answer against another.

At least three. Because ChatGPT gives different answers on different runs, one scan is a single sample rather than a verdict. Your Analysis page has a Last 3 Scans column for exactly this: it shows how many of your three most recent scans found you for each phrase. Appearing in three of three is a consistent result; one of three is within normal variation; none of three means that phrase did not name you in any of them. Until you have three scans it tells you how many more you need rather than guessing. The automatic weekly scan gets you there in three weeks, but your on-demand scans get you there far sooner — running one now, one after you make a change, and one a couple of weeks later is a better set of three than three consecutive days, because the readings are spread across the time your change would actually take effect.

Websites, mostly — businesses' own. We measured the sources ChatGPT cited in our scans to 5 September 2026, across mostly home-technology searches: about 86% were ordinary websites, mostly businesses' own. Directories were about 10%, led by Angi (about 5%) and Houzz (about 3%). Yelp and Google Business Profile were cited zero times; BBB under 1%. Directory listings matter more for Google AI Overview, where Business Profile cards were 12% of citations. For ChatGPT specifically, businesses' own websites are most of what it cites.

There's one thing that overrides all of it. OpenAI states that sites that block its search crawler, OAI-SearchBot, don't appear in ChatGPT search answers. This is worth checking even if you never chose to block anything: a firewall or security service can block a crawler by name without the site owner knowing. indextr checks this on every scan and will tell you if it's the problem.

A current OpenAI model with live web search enabled, so results reflect what the web returns today rather than only what the model learned in training.

It depends on the type of change, and these are our suggestions for when to look, not promises of when something will move.

Google reviews and Google Business Profile updates are worth checking after 2–4 weeks. Website changes are worth checking after 4–8 weeks. Third-party citations — industry publications, local news, directories — are worth checking after 2–3 months.

Those ranges are our judgment, not measured from your results or anyone else’s. No engine publishes a timetable — Google documents that it crawls a site more often when the site changes often, but gives no number, and neither OpenAI nor Anthropic publishes one at all.

Perplexity documents one way it can be faster. Its documentation says that when someone asks it a question it “might visit a web page to help provide an accurate answer” — so a change to your site can show up there sooner than on engines answering from an earlier crawl. Bear in mind Perplexity carries the least weight of the five engines in your score, so movement there lifts your overall score only slightly.

Every recommendation carries the same estimate so you know when it is worth looking again — and if nothing has moved by then, that is information too.

Google Search Central — managing crawl budget · checked 29 August 2026

Perplexity — bots and crawler documentation · checked 29 August 2026

Run an on-demand scan a few days after you make one, and watch Perplexity first — if it is on your plan.

Engines that answer from an index use a copy of your site made whenever their crawler last visited, so a change you publish today stays invisible until they come back, and no vendor publishes when that will be. Perplexity documents that when someone asks it a question it may go and read the page right then. So it can reflect a change sooner.

That makes it an early signal rather than a result. If Perplexity picks your change up, the page is live and a crawler can read it — you can stop worrying about the change itself and wait for the slower engines. If it does not after a couple of scans, check that the page is live and that crawlers can reach it.

It will not move your score much on its own. Perplexity carries the least weight of the five engines in your score. What it gives you is confirmation, early.

Perplexity — bots and crawler documentation · checked 29 August 2026

Start with the recommendation at the top of your list — one to start with, not a verdict that it matters most. Each recommendation shows its priority, and a lower-priority one can be quicker to finish, so you can work on those while a slower one takes effect.

Being mentioned means an AI engine named your business in its response. Being cited means the AI engine included a direct link to your website as a source. You can be mentioned without being cited — the engine named you but linked to other sources, or to none. A citation is also a link a reader can follow to your site.

Two different places.

Some of it is looked up while it answers. The engines indextr scans search the web mid-answer, read what comes back, and summarize it. That part behaves like search: it reflects what is on your site and on other sites, and it can change when you change something and a crawler reads it again. Every recommendation indextr gives you is aimed here, which is why you can act on them and then watch your next scan.

The rest is what the model already knew. Language models absorbed an enormous amount of text while they were built, and they can answer from that memory without looking anything up. That knowledge comes from what had been written about you — press, forums, reference sites, books — and it only changes when a new model is trained, which is not on any schedule you control.

We do not pretend the second one is something you can fix this quarter. The work that moves your visibility is on the first.

It searches first. When someone asks for a recommendation, the engine runs one or more web searches and names businesses from what it finds, mixed with what it already knew.

That is why the sources can differ by kind of business and by engine. In our scans to 5 September 2026, mostly home-technology searches, most citations on every engine were ordinary websites, mostly businesses' own, and directories such as Angi and Houzz were a small, engine-specific share.

The practical rule is the same for everyone: be on the pages that come up when your customers' questions are searched, and make sure those pages describe what you do and where.

It can — and the more useful answer is what we can and cannot measure.

Scans run with live search on, so your score reflects the whole answer, which may blend what the engine just read with what it absorbed during training. Memory only reaches some of the numbers. It can put your name in an answer, but it cannot produce a citation, because a citation needs a real source URL and memory has none. Sentiment, position in the answer and competitive position are all measured only on answers where you appeared, so they inherit whatever caused the mention.

What we cannot tell you is how much of your score comes from memory. It is tempting to assume that being named without being cited proves it, and that is wrong: a citation matches your own domain, so an engine may have found someone else’s page mentioning you and named you from that — which is retrieval, not memory. The two look identical from the outside.

The only clean test is asking with search switched off, which we do not yet run as part of a scan. When we tried it by hand in August 2026, two local businesses were unknown to all three models we tested, while a nationally published methodology was known to one of the three.

So for a business without wide public coverage, our expectation is that memory contributes little and your score is earned mainly by what is findable today — the part you can move.

A little, slowly, and you will not be able to watch it happen.

What a model can say about you without searching was absorbed when it was trained, from what had been written about you — press, trade publications, forums, other people’s case studies.

Allowing the training crawler is a choice, not a lever. It does not make you memorable; it only avoids excluding you. Refusing it costs you nothing in search visibility, because search and training are separate crawlers with separate permissions.

And the payoff lands whenever the next model is trained, which nobody publishes and you cannot observe.

The useful part: most of this work counts twice. A trade write-up or a supplier case study naming you is a page engines can find and cite now, and may feed a future model as a side effect. Do it for the reason you can measure.

Because it isn’t the priority it once was, and we’d rather point you at things with more evidence behind them. Google states plainly that “structured data isn’t required for generative AI search, and there’s no special schema.org markup you need to add.” OpenAI documents one requirement — that its crawler can reach your site.

This is not a reason to remove what you already have. Schema drives Google’s rich results, which is a real benefit that has nothing to do with AI. If it’s on your site, leave it.

What changed is its place on the list, not its value. It is now a maintenance item rather than a project.

Google Search Central — AI features and your website · checked 29 August 2026

OpenAI — crawlers and site owner controls · checked 29 August 2026

It’s a reasonable expectation. Schema markup exists because older search technology couldn’t reliably pick out “this is the phone number, this is the service area, this is the price” from ordinary sentences. Labeling those facts in hidden code solved a real problem.

No engine documents needing that help for its AI answers. We checked all five in August 2026 and none documents a requirement for structured data; Google states there’s no special schema.org markup you need to add.

What indextr measures is where AI engines actually got their answers. In our scans to 5 September 2026, mostly home-technology searches, most citations on every engine were ordinary websites, mostly businesses’ own.

We’d rather be straight with you than sell you a task list. If an engine publishes something that changes this, we’ll change our advice and say so.

Into the visible page. Write the facts where a customer can read them too — your service area, the brands you carry, what you charge, what you don’t do. That’s the same job schema markup was invented for, done in a way that works for the engine and for the person deciding whether to call you.

Put another way: if you have schema, keep it and keep it accurate. If you don’t, we’d put the time into a clear service page before adding markup — it’s the one of the two a customer can read too.

Google ranks pages; AI engines write an answer, drawing on web pages they search at the time, such as business websites and directory listings, and on what they learned in training. A business with strong Google rankings may still not be named in AI answers, which is why it is worth measuring directly.

Every plan gets one automatic scan a week. On top of that you can run scans on demand: Free gets 1 a week, Starter and Pro get 3 (10 in an account's first week), and Done-for-You gets 10.

Weekly rather than daily on purpose. AI answers vary from run to run, so scanning a business that has not changed anything produces a jittery line rather than information. The scan that tells you something is the one you run after you have made a change, which is what the on-demand allowance is for.

Yes, within your weekly allowance — from the Dashboard, Searches or Competitors pages, and that is the scan worth paying attention to, because you run it after changing something. Free gets 1 a week, Starter and Pro get 3 (10 in an account's first week), and Done-for-You gets 10, on top of the weekly automatic scan. The allowance resets for every account at 00:00 UTC on Thursday, and your dashboard shows the reset date.

Upgrading increases the number of AI engines scanned (ChatGPT on Free; ChatGPT, Gemini and Google AI Overview on Starter; all five on Pro), how many scans you can run on demand (1 a week on Free, 3 on Starter and Pro), the number of searches you can track, and the depth of your recommendations. Free accounts get up to 5 brief recommendations per scan. Starter accounts get up to 5 recommendations per scan across all categories, each with step-by-step actions you can follow. Scans pause adding new ones while you still have a dozen open, so the list stays something you can work through. Pro accounts get up to 7, with the same step-by-step detail.

If you cancel your subscription, your account moves to the free tier. All of your scan history, scores, and recommendations are retained — nothing is deleted. If you resubscribe later, your full history will still be there and your trend chart will show data from before and after the gap.

Yes. Email us at support@indextr.ai with the subject line “Delete my account” and we will permanently remove all of your data within 5 business days.