01
Answer engine optimisation, spelled optimization in most of the world, is the work of getting found by an AI assistant: making it name your brand when someone asks it what to buy or use.
Search engines rank pages. Answer engines such as Gemini, Perplexity, ChatGPT and Google AI Overviews recommend a few names and leave the rest out. AEO is about being one of the names, and it rests on two things: the engine being able to read your site, and your site stating plainly the things buyers ask about. The same work is also called generative engine optimisation, or GEO, and the two names mean one thing.
02
SEO earns a position on a list of ten links. AEO earns a mention in a single answer.
In search, a page at position six still gets clicks. In an AI answer, a brand that is not named does not exist for that buyer. AEO also depends on more than your own site: engines read directories, reviews and comparison pages before deciding, so where you are cited matters as much as what you publish.
03
For one of two reasons, and they need different fixes.
Either the engine never read your site before answering, which is a discovery problem solved by being cited on the pages it does read. Or it read your site and still preferred a rival, which is a messaging problem solved by stating on your pages the attributes the rivals were praised for. Oftheard records which of the two happened for every answer, because guessing wrong sends you to rewrite a page nobody was reading.
04
Oftheard measures what AI answer engines say about a brand to buyers in India, and turns it into a plan.
It checks your site with 99 tests to see whether an engine can read and quote it. It puts your buyers’ questions to ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode from India and records who was recommended and why. Then it produces an action plan: the pages to write and the sites to be cited on. Then it measures again, on the same prompts and the same engines, and says what moved.
05
Six: ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode.
The ChatGPT app and the Gemini app are read as a buyer in India sees them, through a licensed data provider, and so are Google’s AI Overviews and AI Mode. Perplexity and Claude are asked through their paid APIs with web search switched on. A prompt pinned to a district goes to OpenAI’s and Google’s own APIs, which can be told the place. Every answer records which engine and which model build produced it, so results are never mixed across engines into one number that describes neither.
06
Not ourselves.
Nothing here logs in to ChatGPT or drives a browser. The ChatGPT app and the Gemini app are read as a buyer in India sees them, through a licensed data provider, and so are Google’s AI Overviews and AI Mode. Perplexity and Claude are asked through their paid APIs with web search switched on. A prompt pinned to a district goes to OpenAI’s and Google’s own APIs, which can be told the place.
07
Because the answers differ by country.
An engine asked from India, in English, cites Indian directories, rupee pricing, Startup India and Indian review sites, and recommends different names than it would to a searcher in the United States. Oftheard fixes the location to India and the language to English for every prompt, so two sessions are comparable and the answer is the one your buyers actually receive.
08
A session puts one collection of prompts to every engine on your plan and stores each answer.
For each answer it keeps who was recommended, what the engine said they were recommended for, which pages the engine read before deciding, and whether your brand was recommended, named only, or absent. The session page compares the result with the session before it, engine by engine and question type by question type.
08a
Yes. Book it for a date and time inside your billing month and it runs on its own, then emails you the answers.
The dialog that starts a session offers two things: run it now, or book it for a time you choose. A booked session needs nobody at the page — it runs on the server exactly as a background run does, and the mail carries the link to what the engines said. Bookings are held against the same monthly allowance as sessions you start yourself, and they are held from the moment you book, not when they run: a plan with eight sessions, two already run and three booked, has three left, and the dialog says so rather than letting the fourth fail overnight. The allowance is counted in sessions, never in prompts or engines — two prompts or fifty, one engine or six, a session is one. A booking has to fall inside the billing month it is charged to, because the allowance resets on the day your plan renews. Cancel one any time before it starts and its session goes straight back into the month.
09
Because the search behind the answer returns different pages each time, and because language models sample their words.
Two identical prompts minutes apart can name different products. That is why a single answer is not a measurement. Oftheard asks many prompts in one session, records the instrument each answer came from, and compares the share of answers that recommend you rather than any one list. There is a longer note on this: why the same AI answers differently every time.
10
As many as your buyers actually ask, in the dozens rather than a handful.
A collection of five prompts cannot hold a stable number. A collection of fifty, across the questions buyers really put to an assistant, gives a share that a few noisy answers cannot swing. Oftheard generates candidate prompts for your brand, sector and buyers, and you keep the ones that are real.
11
Yes, any prompt in a collection: a state, or a district of it.
"Best gold shop near me" means nothing without a place, so any prompt in a collection, generated or typed, can carry one. Pick the state, then the district, and the map shows where the buyer stands. Google AI Overviews, Google AI Mode and Perplexity take the district's centre as a coordinate. ChatGPT takes the district and state as words, since it does not accept a coordinate. Gemini is told the place in words. The same prompt can be added twice with two places, and each is one call, so a place adds nothing to the cost of a session. A prompt with no place of its own is asked from the state on your brief, as every prompt always was. On Google, a "near me" prompt gets the map pack rather than an AI Overview, so the session records the businesses the pack names as Google's answer, marked as such.
11a
Yes. A question can be asked from any state or one of 787 districts, and “near me” questions are read the way a local buyer gets them.
Pick the district when you add a question and every engine is asked from there. For “near me” questions Google often shows a map pack instead of an AI Overview, so the map pack is read as its answer. The site check also covers the local signals: the address and opening hours in the page markup, the Google listing, and whether the pages name the town. A single-location business usually starts with the free look and moves to Starter for the six engines and the plan.
11b
Not yet. Six engines are asked: ChatGPT, Gemini, Claude, Perplexity, Google AI Overviews and Google AI Mode.
Copilot’s answers rest on Bing’s index, so the part you can act on is whether Bing has your pages at all. The site check has a Bing index item that tells you that, and the fix is to submit the site in Bing Webmaster Tools. Copilot will be added as an engine once its answers for Indian buyers can be read reliably.
11c
Not yet. Every question is asked in English, which is how most Indian buyers ask these engines today.
The site check does read pages written in Hindi, Tamil, Bengali, Telugu and the other Indian scripts properly, so a regional-language site gets an honest score rather than being counted as empty. Questions in Indian languages will follow once the English measurement for India is proven.
12
Ninety-nine checks across crawl access, discovery, description, quotability, speed, local presence and agent readiness.
It asks whether AI crawlers are allowed in and actually get the page, whether there is real text rather than script, whether every page describes you the same way, whether a sentence can be lifted out and quoted, and whether the files answer engines look for, such as llms.txt and structured data, are present and correct. On the paid plans it crawls the whole site and reports templates, titles, broken links and duplicate pages. Every check is written out, with why it matters and what to do.
13
No. The site score is a ceiling, not a prediction.
A site an engine cannot read cannot be recommended, so a low score caps you. A high score only means nothing on your side is stopping the engine. Whether it then names you is measured by the sessions, not by the check.
12a
Yes. A small sensor on your site reports every AI crawler visit, page by page.
The site check says whether the door is open; the crawler reads say who walked in. You install a WordPress plugin, a Cloudflare worker or ten lines on your own server, bound to your domain by a site key, and the crawlers page shows which of GPTBot, ClaudeBot, PerplexityBot, Googlebot and the rest read which pages, and whether a page you fixed has been read since. Nothing about human visitors is sent. It is part of the Growth plan, with ninety days of history. The install guide has the steps and a test button that proves it works.
14
It is cheap and it cannot hurt, but it is not what gets you recommended.
An llms.txt file is a plain-text summary of a site for AI crawlers, described in the llms.txt proposal. Some engines read it; none is known to rank on it. Oftheard checks for it because a missing one is a missed chance to state what you are in your own words, and publishes one for its own site at /llms.txt.
15
The major ones do, and many sites block them without meaning to.
A robots.txt rule written years ago to block unknown bots often blocks GPTBot and OAI-SearchBot, ClaudeBot, PerplexityBot and Google-Extended today. Each vendor documents its crawler and what a rule against it switches off. The site check reads your robots.txt and reports which AI crawlers are refused, because a refused crawler cannot read the page that would have got you recommended.
16
What to write, where to be cited, and which prompts are nearest to a recommendation.
The plan joins one session’s answers to your latest site check: the attributes rivals were praised for that your pages do not state, the directories and review sites the engines read that do not list you, the competitors the engines recommend that you had not named, and the failing checks that stand in the way. It is a short ordered list, not a dashboard.
17
A prompt where the engine read your site or named you but did not recommend you.
Those are the cheapest wins: the engine already knows you exist and chose someone else for a stated reason. The plan lists them with the reason, so the fix is usually one page or one paragraph away.
18
Weeks for engines that search live, longer for anything that depends on training data.
Perplexity, Gemini with search and Google AI Overviews read pages at answer time, so a new page or a new listing can change an answer as soon as it is indexed. Repeating a prompt does not teach an engine anything; publishing the page it needs to read does. Oftheard measures the change by running the same collection again on the same engines.
19
Yes, on the paid plans.
Starter checks up to three competitors’ sites and Growth up to ten. Each competitor’s site can be checked once a day, and those checks are not counted against your monthly site checks. Every session also reports the rivals the engines recommend that you had not named, which is usually the more useful list.
20
Four checks of your own site a month, with the scores and the names of every failing and warning check — and one look at the engines.
The look is ten prompts, put to Gemini as a buyer in India, once a month, measured against one competitor you name: who it recommended, what for, and whether it read your site at all. That last part is the whole point, because it tells the two reasons you were left out apart. What free does not include is the reason a check fails or the fix, the other engines, a full panel of prompts, competitor site checks or the action plan. Those are part of Starter and Growth. The full comparison is on the pricing page.
21
Starter is ₹3,999 a month and Growth is ₹9,999 a month, GST extra where it applies.
Starter includes ten whole-site checks a month with every fix, 750 answers a month (three full sessions of fifty prompts on every engine of the plan, or more smaller ones), as many collections as you like, three competitors and the action plan. Growth includes 24 checks, 2,400 answers a month (eight full sessions), ten competitors. Both are billed monthly and can be cancelled any time: sign in, press Request on the pricing page, and the plan is switched on for your account.
22
You do.
Every answer, source and site check stays on your account, is read only by you, and is deleted on request. Nothing from one account is shown to another, and the engines are asked cold, with no context from previous sessions.
23
India, in English, at present.
That is a deliberate scope rather than a limitation of the engines: the value of the measurement is that every prompt is asked the way an Indian buyer asks it, and the plan names Indian directories and review sites. Other markets and languages will follow once the measurement for India is proven.
23a
There is no connector to them. What every plan has is a read-only API and an MCP server, so any tool, script or AI assistant can pull your sessions, plan and sources.
Google Analytics and Search Console connect directly, in Settings: the first shows every visit that arrived from an AI answer, the second imports the searches that showed your site as questions. Semrush’s AI Toolkit and Ahrefs measure AI visibility for a global market; the best-tools page sets out where they and Oftheard differ for a brand selling in India.
23b
Yes. Both mean how often a large language model names or recommends your brand when a buyer asks it something in your category.
Some tools call it AI share of voice, generative engine optimisation or answer engine optimisation. Oftheard calls the number Share of Model: of the answers the engines gave to your buyers’ questions, the share that recommended you, measured per engine, with the pages each engine read before it answered. AEO vs GEO explains the terms.