Frequently asked
questions

A site URL is enough to begin. You do not need to prepare materials in advance — answer engines judge from public information too, so auditing without internal documents is in fact closer to the real state. Three things do speed the audit up if you send them: two or three brands you regard as competitors, the questions or keywords you would like to be recommended for in AI answers, and a list of the channels you run besides the website (blog, social, marketplaces).

It is fine, but we check first, because the range of what can be changed differs per platform. The key checks are whether structured data can be inserted directly, whether titles and descriptions can be set per page, and whether body text lands in the HTML source. Some builders allow all three; on others the body is painted in by JavaScript and looks like an empty page to a crawler. In the latter case content will not be cited however well it is written, so we tell you that first and decide the alternative together.

It is workable, but we state the limits first. To be cited, an answer engine has to be able to read the document, and Naver Blog blocks major AI crawlers including GPTBot, PerplexityBot and ClaudeBot in its robots.txt. So blog posts remain valid for Naver search but carry almost no weight as evidence for overseas answer engines. With channels alone, what can be done is limited to roughly tidying up brand notation. We say plainly that having your own domain is better over the medium term — and whether that is the investment to make now is a judgement we reach together, with the audit results in hand.

It can. Many clients run paid media or SEO elsewhere and hand us only AEO. One condition, though — if two agencies touch the site's technical settings (robots.txt, structured data, canonical URLs) at the same time, they overwrite each other's work. So at the outset we divide in writing who owns which area. Navirang keeps to the boundary drawn, and problems found outside it are reported rather than fixed.

It runs remotely. AEO work centres on measurement, documents and structural change, so there is almost no stage that requires meeting in person. We have a head office in Gwangju and a branch in Seoul, which makes in-person meetings easy in those two regions; elsewhere we work by video call and document. Kickoff and report walkthroughs happen on video, and communication in between runs through an agreed channel.

Four steps. First, we reply within one business day. Second, we confirm the site URL and competitor brands to set the audit scope (no documents needed to this point). Third, we run the free audit over one to two weeks and deliver the report. Fourth, with the report in hand we settle the six lines that move a quote and give you a figure. The audit report is yours regardless of whether you sign, so stopping after the third step and proceeding internally is a legitimate outcome.

Yes. Measurement is entirely about holding the conditions constant, and when the person changes the judgement criteria shift subtly, which is itself noise in the data. So whoever designed the question set also runs the re-measurement. To keep that from becoming knowledge held by one person, the question set, the measurement conditions and the adjudication criteria are all documented, and the client holds those documents too.

Monthly, under monthly operations. The report carries the raw observations alongside the summary figures — which question was asked, when, under what session conditions, and what came back. A report with only summaries cannot be verified. On a weekly measurement cadence, weekly observation records accumulate and the interpretation is consolidated monthly.

We will. The improvement priorities in the audit report mark the items you can handle internally. Unifying business name and address notation across channels, or clearing out old pages, is usually faster and cheaper done in-house. Not tying work you do not need to hand over into a contract is a principle here.

By putting the agreed question set to the engines again under the same conditions. Three things are held fixed — the question wording, a logged-out session (because personalization moves the result), and the same day and time of day. The results are then counted as mentions and citations separately: the brand name in the body of the answer is a mention, our domain in the source list is a citation. The two move independently, so counting them together leaves you unable to tell what to fix. The report includes the raw observations alongside the aggregates, so you can see for yourself where each number came from.

Hard to conclude that. Search rankings tend to let an established document hold its place for a long time, but answer engines re-select the documents they will use as evidence for each question at that moment, so the slot is not fixed. What is true is that building a state where a brand is confirmed repeatedly across several sources takes time. The precise answer comes from the audit — whether what is cited for those questions today is a competitor's site, a community post or a government document changes the difficulty completely. Pressure of the 'start now or it's too late' kind is a sales line, not evidence.

It will not. Most of what AEO does runs in the same direction SEO already recommends — keeping body text in the HTML source, making structure explicit, marking facts up as structured data, and not blocking crawlers. One thing does genuinely change: rewriting paragraphs into units that stay complete when cut alters the rhythm of the writing. Reviewing that so it does not get worse for human readers is part of the work, and we track whether rankings move alongside.

We do, and we publish the record along with the conditions. We measure 40 questions × 7 answer engines = 280 cells as a baseline against our own site and publish eight weeks of change as a series. The full question set, the measurement protocol, the denominator for the citation rate and the raw CSV are all in scope. Everything we measure about ourselves gets published — because we have asked others, in our own articles, to accept that a figure without measurement conditions cannot be verified, and the same standard has to apply to our own. Measurements of client sites, by contrast, are not published.

"Ranking" does not quite apply. An AI answer has no first or second place the way a search results page does, so there is no position to ask about. What is measured instead is whether the brand is mentioned in the answer, whether the site is cited as a source, and how often that happens when the same question is asked repeatedly. "ChatGPT exposure", "AI exposure" and "AI search exposure" all point at the same work; the precise names for it are AEO (answer engine optimization) and GEO (generative engine optimization).

There is an order. First, check that ChatGPT's search crawler (OAI-SearchBot) can read the site — if robots.txt or a firewall blocks it, nothing after that matters. Second, check that a sentence answering the question exists in the HTML source. If the body is painted in by JavaScript, a crawler sees an empty page. Third, check that the company name, address and category are written the same way across every channel. Fourth, look for documents about you that live outside your own site. Most sites are stuck in the first three.

For questions of the form "recommend a good X", the shortlist usually comes from third-party documents that list several companies against the same criteria — not from any single company's own page. Your own site tends to fill in the description after you are already on the list. That is why fixing only the website rarely gets you into recommendation answers: it takes both the inside of the site (so machines can read what you do) and the outside (documents that treat you on the same terms as others).

There is no business listing to register for. OpenAI's own publisher FAQ answers this question with a single line — "Any public website can appear in ChatGPT search" — with no application and no review. What does exist are free listings on the sources that AI reads: Naver SmartPlace and Search Advisor, Google Business Profile, Bing Webmaster Tools and Bing Places. If someone offers to "register you with ChatGPT", ask which registry they mean.

For an offline business the order differs slightly. Start by making the name, address, opening hours and category consistent between Naver SmartPlace and Google Business Profile — if the details differ per channel, machines cannot tell it is the same shop. Then check that the website is open to crawlers, and that a sentence somewhere answers what customers actually ask ("a good X near Y"). The first two are free and you can do them yourself.

The groundwork mostly overlaps; what differs is the control point and the material each engine reads. ChatGPT's visibility is governed by its search bot (OAI-SearchBot); for Gemini the Google-Extended token in robots.txt covers a different scope; Naver's AI search reads place data, visitor reviews and blog posts together for location questions. So making a site machine-readable carries across engines at once, but verification has to be done engine by engine.