Before a buyer visits your site, they now ask an assistant about you. So does a candidate weighing an offer, a journalist deciding whether you are worth a call, and an investor sizing up your category. They type your name, or your category, into ChatGPT, Gemini, Perplexity or a Google AI Overview, and they read the paragraph that comes back as if it were fact. That paragraph is your new front page. You did not write it, you cannot see every version of it, and for most brands it is being written right now with whatever the open web happens to say.
This is a reputation surface, not a search ranking. A ranking is a list you can scroll past. An AI answer is a verdict delivered in one confident paragraph, often with no link clicked at all. If your brand is described wrongly, described thinly, or left out while a competitor is named, the decision is shaped before you get a word in. The good news is that these systems are not arbitrary. They draw on sources, and sources can be shaped, corrected and corroborated. That is the work.
Why AI answers decide what they say about you
It helps to understand roughly how an assistant arrives at a claim about a brand, because the method tells you where to intervene. There is no single dial to turn. There are a few inputs that all point in the same direction, and your job is to make them agree.
- Training data. The model has read a large slice of the public web up to a point in time. What was written about you, and how consistently, sits inside that baseline understanding.
- Live retrieval. Many assistants now fetch current pages at the moment of the question. Your own site, your profiles and recent coverage can all be pulled in and quoted, which is why fresh, clear pages matter.
- Trusted sources. These systems lean harder on places they judge credible: established press, reference sites, and structured directories. A claim carries more weight when it comes from somewhere the model already trusts.
- Corroboration. A fact that appears in only one place is fragile. A fact repeated across your site, the press and a reference entry looks settled, and settled facts are the ones an assistant is comfortable repeating.
Read that list back and the pattern is obvious. Assistants reward brands that say the same true things about themselves in many credible places. They punish inconsistency and silence, not by penalising you, but by filling the gap with a guess or a rival.
Search optimisation and answer optimisation are not the same job
Search engine optimisation aims to rank a page in a list of links, so a person clicks through and reads it. Answer engine optimisation aims to have your facts extracted and attributed inside the answer itself, so the assistant quotes you whether or not anyone clicks. The two overlap, and they reinforce each other, but the target is different. One wins a position on a page. The other wins a sentence in a paragraph that stands in for the page.
That difference changes how you write. For search you can bury the answer three scrolls down and still rank. For answers you cannot. If a machine has to hunt through your prose to reconstruct what you mean, it will more often quote the source that stated it plainly. The clearest, most quotable version of a fact tends to win.
A search ranking is a list you can scroll past. An AI answer is a verdict in one paragraph. If your brand is missing from it, the decision is made before you speak.
The practical method
The work divides into four moves that compound. None of them is a trick, and that is the point. You are making it easy and safe for a system to describe you correctly, then earning the outside corroboration that makes the description stick.
Write answer-first content
Open every important page with a short, self-contained answer to a real question, phrased the way a person or a model would ask it. Aim for a passage of roughly 40 to 60 words that could be lifted and quoted without the surrounding context. Then let the rest of the page support it. One clear answer per section, in plain language, beats a clever paragraph a machine cannot parse.
Structure the page for machines
Add the structured data that assistants and search features rely on: Organization, FAQPage, HowTo and definition markup, plus clean headings, lists and tables. Structure is not decoration. It tells a system which sentence is the answer, which words are the entity, and how confident it can be that the two belong together.
Keep your entity facts consistent
Decide the handful of facts that define your brand, what you do, who you serve, where you operate, when you started, and state them identically everywhere: your site, your profiles, your press kit, the open web. An assistant that sees the same founding year, the same category and the same locations in five places treats them as fact. One that sees three different descriptions hedges, or picks the wrong one.
Earn citations that corroborate you
The strongest signal is a credible third party saying what you say. Earned coverage, reference entries and mentions from sources the models weigh heavily do more than any on-page tweak, because they turn your claim into a corroborated one. This is where reputation work and answer visibility become the same discipline rather than two separate budgets.
How to audit what the assistants say about you
You cannot fix what you have not read. Start by asking the assistants directly, the way your buyers would, and write down what comes back.
- Ask each major assistant about your brand by name, then about your category without naming yourself, and note whether you appear, how you are described, and who is named alongside you.
- Record the specifics the answer gets right, gets wrong, or leaves out entirely, and capture the sources it cites so you can see where the claim came from.
- Repeat the priority questions your buyers, candidates and investors actually ask, not just the flattering ones.
- Trace every wrong or missing fact back to its source on the open web, because that source, not the model, is what you will correct.
Run this on a schedule, because the answers move as the web moves. A brand that looked absent this quarter can appear next quarter once corroboration accumulates, and the reverse is also true.
What to do when an assistant gets you wrong
The instinct when an answer is wrong or outdated is to find a way to talk to the model. There is no such lever, and the vendors that claim one are selling a story. Assistants describe you from the sources they can see, so you correct the sources, not the model. If your own page states the outdated fact, fix the page. If a profile or a reference entry is wrong, correct it there. If the press repeated something stale, earn newer, accurate coverage that gives the system a fresher and more credible thing to quote.
Why nobody can guarantee a placement
These systems are not ours to control. They change their models, change what they retrieve, and change how they weigh sources, often without notice. Anyone who promises your brand will appear in a specific answer is either misunderstanding how the systems work or hoping you do. Treat a guaranteed placement the way you would treat a guaranteed stock return. What is honest, and what actually works, is to build the conditions that make an accurate citation far more likely, then measure it and improve.
This is the same stance we take across everything we publish. We will tell you where you are cited and where you are not, including the answers that still leave you out. That candour is the point, because a source people can trust is exactly the kind of source an assistant learns to quote.
Measuring AI visibility honestly
Honest measurement tracks how often and how accurately the assistants mention you across a fixed set of priority questions, which sources they cite alongside you, and how that picture changes over time. Pair it with your search and coverage data so you see the whole reputation surface, not one vanity number. The goal is not a single screenshot of a good answer. It is a trend: more of your priority questions returning an accurate, attributed mention, quarter after quarter, and fewer that hand the moment to someone else.
You will not control the output. You will steer the inputs, read the results plainly, and keep closing the gap one question at a time. That is the whole method, run from our Kolkata and Mumbai offices as one program across the press, search and AI answers.