Frequently asked questions

Services

How long does GEO take to show results?

Access fixes — robots.txt and llms.txt — take effect on the engine's next crawl, usually within days. Content and markup work returning as citations depends on how often that engine refreshes its index. Galata Media takes the baseline measurement in week one, so progress is read from measurement rather than estimated.

Does our existing SEO work go to waste?

No. Technical health, crawl access and topical authority do the same job on both sides. What changes is the shape of the content: a generative engine quotes a single passage rather than the whole page, so that passage has to stand up alone. Galata Media restructures existing pages around passages instead of rewriting them from scratch.

How do you verify that we are being cited?

An agreed set of questions is asked in each engine at regular intervals, and we record whether the brand appears as a source in the answer. Measurement is per engine because the page ChatGPT cites and the page Perplexity cites are frequently not the same one. The report shows questions won and lost together.

How much does a corporate website cost?

Three variables set the cost: how many pages, how many languages, and whether the content already exists. Galata Media quotes after the page map is drawn, because a number given before the map is a guess and usually breaks as scope grows. There is no published package list; each scope is priced on its own.

Do you build on WordPress?

No. Corporate sites are built on a statically compiled stack; pages are not regenerated on the server for every request. That has two practical consequences: load time stops depending on how many plugins are installed, and the security surface an unpatched plugin opens disappears. Content updating is handed over as a separate workflow.

Can you build an identity around our existing logo?

Yes, and it is often the right call. If the existing mark is recognised, replacing it resets the recognition you already paid for. In that case the mark is not redrawn: its geometry is measured from the original file, and only the missing rules — scale, clear space, colour — are completed. If a change is needed, the reason is written down first.

Which e-commerce platforms do you work with?

The platform is an input to the consulting, not its subject: we work on whatever you already run. Most of the work — product data, markup, feeds and copy — is platform-independent. If a migration is genuinely needed we raise it as its own decision with its own reasoning, rather than smuggling a replatforming project inside a consulting engagement.

What share of the ad budget does the agency take?

The management fee is a fixed monthly amount, not a percentage of spend. A percentage model rewards the agency for growing the budget, when in most accounts the right move is not a bigger budget but cutting the search terms that waste it. A fixed fee removes that conflict of interest.

What sets the price of social media management?

Three variables set it: how many channels, how many pieces of content a month, and whether production is included. Without shooting the cost drops noticeably, but the content is then limited to imagery the brand already owns. Galata Media prices those three separately rather than quoting one package figure.

Tools

Can the audit harm my site?

No. The audit sends one GET request to three addresses: the homepage, /robots.txt and /llms.txt. It signs in to nothing, submits no forms, writes nothing and leaves no trace on the site. The requests identify themselves as GalataMediaAudit, so that is how they appear in your server log.

What score should I be aiming for?

The number is not the target; which criterion fell is. Crawler access carries the heaviest weight at 24 points, because points lost there make the other nine meaningless — a page that cannot be read has neither its markup nor its answer passage evaluated at all. Fix access first, then move to passage structure.

Should I block AI crawlers?

These are two separate decisions and should not be confused. Blocking training crawlers (GPTBot, Google-Extended, ClaudeBot) limits your content being used to train models, and is a legitimate licensing choice. Blocking the crawlers that produce answers (OAI-SearchBot, PerplexityBot, Claude-SearchBot) removes your brand from those answers entirely.

Guides

I have no budget — where do I start?

Two jobs cost nothing and are the highest-return ones: writing down the questions your customers ask, and verifying that your site can be crawled. The first is an hour with your sales team, the second a check you can run with free tools. The third step — writing a page for each question — takes time but still not budget.

Do customers come from social media?

They do, but it does a different job: social creates demand, search captures the demand that exists. If your product sells on looks and people do not yet know to search for it, social comes first; if they are already searching, building the search side first is cheaper. The two are not alternatives — their order differs.

How long before demand starts arriving without ads?

For a business whose site is crawlable and whose pages are written, the first movement usually shows in four to twelve weeks; longer on contested queries. What matters here is that month one's output is a record rather than traffic: without measuring where you stand on which query, you cannot say what changed three months later.

Can I pay ChatGPT to appear?

For organic answers there is no such channel: there is no submission form and no directory, and visibility depends entirely on crawler access and page structure. A service sold as "we get you added to ChatGPT" should not be bought — the only sellable work is opening access and making the content quotable.

Can traffic from this channel be measured?

Partly. When a user clicks the link in an answer it shows in analytics as a referral from chatgpt.com or perplexity.ai; someone who merely reads your name and then contacts you directly cannot be attributed. So measurement runs on two legs: the referral in analytics, and a regular test recording whether the brand appears in the answers at all.

Is organic traffic free?

Nothing is paid per click, but it is not free: writing the pages, the technical fixes and the measurement all take work. The difference is where the money goes — in advertising it goes to the click and ends when you stop; in organic it goes to work done once that keeps working in year two. Over time the unit cost falls.

Will AI summaries kill organic traffic?

It lowers the click rate on informational queries, while on commercially intended ones users still go to the source — the decision is made on the site. The right response is not withdrawing content but widening the target: earning the click and being cited inside the summary. Both are fed by the same writing.

Do free directory listings help?

For most, no. If your sector genuinely uses two or three directories, listing there with correct details is worthwhile; bulk-registering in dozens of general ones brings neither visitors nor visibility. Worse, addresses and phone numbers written differently across directories make it harder to resolve the brand as one entity — so it is not merely useless, it can do harm.

How long does it take with no budget?

Access fixes and the business profile are measured in days; page writing reaching a result forms a band of four to twelve weeks in practice. That is exactly where advertising has the advantage: it starts today. Which is why a business with tight cash flow runs both in parallel — ads fund today, visibility work funds next year.

What is the difference between SEO and GEO?

SEO works for a position; GEO works for being inside an answer. The overlap is large: both require the page to be crawlable, to answer the question cleanly, and the brand to be resolvable. Where they part is measurement: SEO measures rank and clicks, GEO measures whether the brand name appears in the answer.

Can I do SEO myself?

Most of the first strand, yes: checking robots.txt, clearing noindex, setting up Search Console and submitting a sitemap can all be done with free tools, and they produce the fastest gains. It gets harder at the third and fourth strands — sustained content production and making the brand record consistent externally; those need time and continuity.

How long does it take to rank higher?

A proposal giving a firm calendar is promising something it does not control. What is measurable: technical access fixes show their effect within days, while content and structure work reaching the point of being recrawled and reassessed forms a band of four to twelve weeks in practice. On competitive queries that band stretches.

Does buying ads lift my organic ranking?

No, and the reverse holds too: stopping ads does not lower organic position. They are separate systems. There is an indirect link, though — visitors arriving via ads learn the brand and later search for it by name, which raises organic traffic. That is not the same as buying a position.

What are firms offering ranking guarantees selling?

Usually one of two things: first place on a long phrase nobody searches for and nobody competes on — technically true, commercially worthless — or the advertising slot. When reading a guarantee, look at which query and which position it covers; if the query is unnamed, the commitment cannot be measured.

How long until a new site gets indexed?

Anywhere from a few days to a few weeks; there is no single timetable. Three things speed it up: verifying the property in Search Console, submitting the sitemap, and having at least one external link pointing at the site. What slows it down is a page with no inbound link at all — the crawler needs a route to discover that address.

Search Console says 'Indexed' but I cannot find it in search

Being indexed and being on the first page for a query are different things. A site: search confirms the page is in the index; if you cannot find it in search, the issue is ranking, which is content work. Check which query you tried, too: searching your own exact title is not a ranking test.

I have no physical shop — can I still add it?

Yes. Businesses that go to the customer can open a profile by defining a service area; the address is then hidden and the profile shows the regions served instead. To avoid leaving a home address exposed, set this during verification rather than after.

Does adding keywords to the business name help?

It may look like a short-term gain, but it is a policy violation and can end in the profile being suspended. More importantly, the moment the name on the profile differs from the one on your website, entity resolution breaks: the engine sees two different names. The right route is filling in the category and the service descriptions properly.

Do reviews affect ranking?

In local results the number and recency of reviews is a signal, but the real effect is on clicks: with two profiles side by side, users go to the one with reviews. Buying fake reviews works the other way and can bring enforcement against the profile. The sustainable route is building a request for a review into the process after delivery.

How long does a corporate website take?

Duration is set not by page count but by two things: how many external systems are involved and when the content is ready. At Galata Media a corporate site with no integrations and content supplied up front generally runs six to ten weeks from design to launch — and the item everyone waits on is almost always copy and imagery, not code.

Where does the price gap between a template and a custom design go?

The gap does not go into looks but into three things: page structure built around your actual content, the technical work for speed and accessibility done from the start, and the ability to add new page types as you grow. A template has decided all three in advance, independently of your content; where it fits, it is very reasonable, and where it does not, it produces cost on every new page.

How should payment be staged?

Tying payment to deliverables rather than to dates protects both sides: scope sign-off, design sign-off, end of development, launch. Date-based payment runs the same way regardless of who caused a delay, and it produces arguments on projects waiting for content. Leaving the last instalment until after launch is what makes handover actually complete.

Is refreshing my current site cheaper than a new one?

Sometimes, but less often than assumed. Working on top of an old system means buying its limits as well; on a platform that has not been updated in three years even a small change takes longer than expected. The test is not age: how many days does adding a new page type take, are security updates still shipping, and can the content team publish a page without asking you.

Can I copy a site I like outright?

Copying a design is both legally risky and ineffective, because that site's structure was built around its own sales process. What can be copied is the decision: why they explain things in that order, why that page exists, why the pricing approach sits where it does. Adapted to your business, the result does not look like theirs — and should not.

Why are your own references kept out of the index?

Because a client's brand should not be tied to our domain in search. The reference pages exist on the site and can be reviewed; what we prevent is those pages surfacing under our name on the client's own brand queries. If you want to see an example, we show live addresses in the conversation — the same evidence, without occupying the client's search results.

Logo first or website first?

If there is no brand identity, that comes first: the site is the first surface the identity is applied to, and a colour or typeface changed later means reworking the whole site. If an identity exists, the order does not matter. When Galata Media takes both together, identity decisions get tested on the site templates — which removes the risk of a system that looks good on paper and falls apart on screen.

How many pages should it have?

Page count is an outcome, not a target. The right question is whether every question a client asks before deciding has a page. Services, the pricing approach, references, team, contact and frequently asked — on most corporate sites those six are the scope. Adding pages later is easy; splitting one enormous page that should have been six is not.

Should the agency register the domain?

No. The domain should be registered in the company's name using the company's own email; granting the agency management access is enough. A domain sitting in an agency's account becomes an asset that has to be transferred when the relationship ends — and sometimes becomes a bargaining chip. The same rule covers hosting, analytics and Search Console.

What do I need to do after launch?

Three jobs, all in the first week: verify the property in Search Console and submit the sitemap; if this was a rebuild, check one by one that the old addresses land on the new ones; and confirm analytics is actually collecting. Skip those three and the problem surfaces months later, as a traffic drop.

Is WordPress enough for a corporate site?

For most corporate sites, yes. Where it falls short is when page types change often and the content is structural: if every service has different fields, if references need four separate templates, the data model starts being assembled from plugin on plugin and the maintenance cost overtakes that of a custom build.

Does buying a theme make sense?

If the budget is tight and your content structure fits the layouts the theme offers, yes. The risk: purchased themes carry a great many features for every eventuality, and that code loads whether used or not. The question when choosing one is not how good the design is but how many plugins it depends on and how much of the page weight comes from features you will never use.

Can I move off WordPress later?

Yes, but the cost depends on how the content is stored. If it sits in structured fields, moving is a data export; if it was produced by a page-builder plugin, markup and text are tangled together and every page has to be cleaned by hand. That is the part of a plugin decision that gets billed three years later, not today.

What PageSpeed score should I aim for?

The score is not a target. The target is field data showing LCP under 2.5 seconds, INP under 200 milliseconds and CLS under 0.1. Meet those three and the score rises on its own; chasing the score leads to cosmetic changes that improve none of the three things being measured.

Is switching hosting enough to get faster?

If server response time is genuinely slow it makes a difference, but on most sites the biggest delay happens in the browser, not on the server: large images, late-loading fonts and third-party scripts. Measuring server response time before committing to a move prevents a wasted migration project.

Does a cookie banner affect speed?

Yes, and usually in two places at once. When the banner loads late and pushes the page down, CLS breaks; and because most banner tools load their own script, it weighs on INP too. The fix is not removing the banner: reserving the space and taking the tool's script off the critical loading path is enough.

How many revisions should I get?

Revision count is the wrong measure; the presentation structure matters more. Once three directions are shown and one is chosen, the revisions that follow stay at the detail level and their number stops mattering. A promise of unlimited revisions usually means the direction will never be settled. At Galata Media the flow is three presentations: discovery, direction, application.

Isn't an AI-generated logo enough?

Enough to produce an image, not enough for a brand mark — and the gap sharpens on the legal side. Verifying that a generated image is original and does not resemble an existing registered mark is still human work; a mark refused at registration costs far more than the generation did. Clean vectors, legibility at minimum size and a single-colour version are not in the output either.

What is the difference between a logo and a brand identity?

A logo is a mark; a brand identity is the set of decisions around it: a colour system, typography, a photographic language, layout rules and the manual that records them. In companies that buy only a logo, the typical outcome within a year is every department picking its own colour and typeface — and recognition is the work of consistency, not of the mark.

How long does registration take?

The timeline is not a single stage: examination, publication in the Bulletin and the opposition window run in sequence, and the total shifts markedly depending on whether an opposition arrives. Current durations should be followed from TÜRKPATENT's own information pages; an agency promising a date would be overstepping, because the institution sets the clock.

How many classes should I file in?

The classes covering what you actually sell today, plus the areas you plan to enter within three years. Adding classes "just in case" raises the fee and can create exposure to cancellation requests in classes you never use. The right approach is to put the class list next to your business plan and decide against it.

Is using an unregistered logo a problem?

You may use it, but the protection is weak. The common scenario with an unregistered mark is that another company registers the same or a similar sign in your sector, after which the argument about who may use it runs against you. For companies raising investment, appointing dealers or exporting, registration usually turns up as a requirement.

Is selling on a marketplace cheaper than my own site?

At the start yes, at scale no. A marketplace takes setup cost close to zero but charges commission on every sale and keeps the customer relationship. Your own site does the opposite: costly up front, but repeat purchase and customer data stay with you. For most brands the answer is running both — the marketplace as an acquisition channel, the site as the loyalty channel.

How many products should I launch with?

As many as you can enter with complete data and good photography. Publishing thirty products fully beats publishing three hundred half-filled, both for sales and for visibility: incomplete data stops a product from being recommended and from being approved in ad channels.

When should the payment gateway application go in?

Before development. The application is administrative rather than technical: company documents, a description of the business and sometimes a turnover declaration are requested, and assessment can take days. Waiting on payments once the site is ready is the most common and most easily avoided delay.

What happens if the search work does not deliver?

With an exit clause and a transfer clause in the contract, parting is clean: the accounts stay yours, the published content stays on your site, the measurement history is in your hands. Without those clauses, leaving can also mean losing the content and the data. The right question is not "do you guarantee results" but "what do I keep when we part".

How long before results show?

Technical access fixes pay off fastest, measured in days. Content and structure work reaching results depends on pages being recrawled and reprocessed; in practice the band runs four to twelve weeks. A proposal offering a firmer calendar than that is promising something it does not control.

Monthly retainer or project-based?

Technical fixes and the initial build are finite work that suits a project fee. Content production and measurement need continuity and fit a monthly arrangement. The common mistake is folding both into one monthly package until what was done in which month becomes invisible; separating the two makes what you are paying for visible.

Is freelance work lower quality?

No, and the question is on the wrong axis. Individual skill has the same distribution on both sides; what differs is process and continuity. On a job solved by a single skill, an experienced freelancer beats an agency without a process. On a multi-part job, the structure that joins the parts is what decides.

What if the freelancer disappears?

What removes the risk is not the contract but the access structure. If the accounts are in your name, the source files live in your cloud storage and the repository sits in your company account, changing people slows the work down rather than stopping it. Without those three, the relationship ending becomes the same event as the project ending.

Is it a problem if the agency opens the account in its own manager account?

Yes, and the cost is paid at the moment you part. In an ad account opened inside an agency's own manager account, the campaign history, conversion data and negative keyword list stay with the agency; the next agency starts from zero and the algorithm's learning restarts. The right structure: the account in your name, the agency granted manager access, and that access revoked when the relationship ends.

Could I not run it myself?

On an account that starts with one campaign, a clear product and limited competition, yes — and that is the right way to start. An agency earns its fee once the account spreads across several campaign types, the search terms list needs weekly maintenance and landing pages need testing. The test is not the size of the budget but how many separate decisions the account needs each week.

Could I just keep the address structure?

You can, and where it is possible that is the safest route. Keeping the address structure removes the redirect risk entirely. Changing it makes sense only when the existing structure is genuinely a problem — unreadable parameter URLs, no language segmentation, addresses written from IDs instead of names — and in that case the 301 table becomes mandatory.

How long does a rebuild take?

Duration is set not by page count but by how the content moves. If existing content is carried across as-is, it is shorter than building new; if the content is rewritten, it is longer, because copy becomes the bottleneck. At Galata Media a rebuild with no integrations runs six to ten weeks when the content is ready.

Can you run an online store as a sole trader?

Yes, and it is a common way to start. The decision point is tax and liability: as turnover grows and outside investment or a partner appears, a limited company structure becomes necessary. Some banks also ask for different documents depending on company type during the gateway application — which is why the company decision belongs before the application.

Should I start on a marketplace first?

It is a reasonable way to test demand: setup costs little and you find out quickly whether the product sells. But what you accumulate on a marketplace is turnover; the customer relationship and the data do not stay with you. The usual right order is: validate demand on the marketplace, build your own store from the same product data, then run both.

Should the ad budget be inside the agency fee?

No. They belong on separate lines, because one is a fee for work and the other is money paid to a platform. Merged into one figure, how much went to advertising becomes invisible and it stops being clear who benefits from a budget increase. Ad spend should go directly to the platform from the company's own account.

Can we produce the content and the agency just post it?

Yes, and in some companies it is the most efficient arrangement — particularly where production capacity already exists in-house. What you are buying then is planning, the calendar and community management, and the price should fall accordingly. The one thing to watch is whether production is genuinely sustainable: when the content flow stops in month two, the plan stops working with it.

What if the agency refuses to hand over the files?

The contract decides. If transfer of intellectual property and delivery of source files are written in, the request has a legal basis; if not, the obligation becomes arguable. That is why those two clauses belong in the contract while the relationship is good — they cannot be added once it is not.

Can an ad account be transferred?

It depends whose name the account is in. If it is the company's own account, the handover is a change of manager access and the history is preserved. If it is a sub-account created inside the agency's manager account, the transfer depends on their consent; without it you open a new account and restart the learning period.

Can you start with a small budget?

You can, but a very small daily budget does the opposite of what is expected: the campaign never gathers enough data to learn which searches work, and unit cost stays high. The more efficient route is to concentrate the budget on one product and one region rather than spreading it across the whole country and catalogue.

How much of the budget should go to brand versus sales?

Until the measurable sales channels are saturated, it is more sensible not to split. Allocating budget to a brand campaign while there is still unmet demand on buying-intent searches moves measurable profit into an unmeasurable line. Once the sales channel saturates — that is, once raising budget starts raising unit cost — moving to the brand side makes sense.

How do you compare results across the two?

Comparing on a single measure misleads. Search ads catch someone already at the decision and look cheaper at first sale; social ads bring someone who has not decided yet, so they look expensive at first sale and can be strong on repeat purchase. The sound comparison is on the six-month value of the customer acquired.

Which is better for retargeting?

The social side is usually more flexible here, thanks to the range of visual formats and frequency of display; but on either platform retargeting only works with measurement correctly installed on the site. Without it, retargeting cannot tell someone who abandoned a cart from someone who saw the page for three seconds, and the budget goes to waste.

What if the agency insists on its own contract?

That is normal and not a problem; the problem is if the six clauses are missing from it. Offering the missing clauses as an addendum to an agency's own contract is a common and accepted route. If an agency refuses to add the account-ownership or IP-transfer clause, that refusal is itself the information you were looking for.

What is the difference between transferring IP and granting a licence?

In a transfer the right becomes yours; in a licence it stays with the agency and you are granted use under stated conditions. The difference shows up at trademark registration, when you want another agency to build on the work, and at a sale of the company. On logo and identity work, having the transfer in writing matters particularly.

Can you migrate later?

Yes, and the cost depends on how disciplined the product data is. A catalogue with its fields properly filled moves as a data export; a catalogue whose descriptions carry embedded markup has to be cleaned product by product. Preserving the address structure and preparing a 301 table are inseparable parts of the move.

Why does exporting a marketplace feed matter?

Because the store does not stay the only sales channel. Sending the same product data to marketplaces and ad channels with the right fields is the platform requirement people notice latest. On a platform that cannot produce feeds, the catalogue gets re-entered by hand for every channel — and as you grow, that becomes a full-time job.

Should we keep the raw footage?

Yes, and it needs to be in the contract. If the shoot was paid for from your budget, the raw frames are your asset; being able to reuse the same material after changing agency depends on it. When only the selected, edited frames are handed over, two years later all you hold is those posts.

Is producing in-house cheaper?

With continuity yes, without it no. In-house production pays off by paying for equipment and learning once and using them every month; set up for a few posts a month, the equipment and the time sit idle. The common answer is hybrid: routine product frames in-house, campaign and brand shoots outside.

How many options should a presentation show?

Fewer is a good sign. Three reasoned directions are more useful than fifteen variations: a long list means the designer is not defending a direction and is leaving the decision to you. In a good presentation each direction explains why it is that direction, and the choice becomes strategic rather than aesthetic.

Is looking like our competitors a problem?

Yes, for two separate reasons. The commercial one: recognition is built on difference, and a mark repeating the sector's clichés is not remembered. The legal one: closeness to a registered mark in the same classes can draw an opposition to your application. Which is why looking at the sector's existing marks is work to be done before the presentation.

How long should a brand guideline be?

Page count is not a target; the test is whether a printer or a new hire can do the work without asking a question. In a small company that usually lands at 15–25 pages; multi-brand, multi-country structures run longer. The common failure of hundred-page guidelines is mixing applicable decisions with brand storytelling in the same document.

Should the guideline be delivered as a PDF?

A PDF is practical for distribution but not sufficient on its own, because it cannot be updated and the files inside it cannot be extracted. Proper delivery has three parts: a readable PDF, an editable source document, and a folder holding every logo and colour file the guideline refers to. Without all three, the guideline drifts out of step with the files actually in use within six months.

What happens if there is no robots.txt file?

With no file and a 404 at that address, crawlers assume there are no restrictions and crawl freely. For most corporate sites that is harmless; the problem is that nobody can tell whether the absence is a decision or an unfinished task. Publishing an empty file — just User-agent: * and an empty Disallow line — makes the decision legible.

If I block GPTBot, do I disappear from ChatGPT entirely?

No. GPTBot is only for model training. When ChatGPT searches to answer a question it uses OAI-SearchBot, and when a user opens a link it uses ChatGPT-User. With those two open you can appear as a source in answers while GPTBot stays blocked — and for many companies that is exactly the right configuration.

Does the Crawl-delay directive work?

It depends on the crawler, which is why it is not a reliable rate control. Anthropic's documentation gives a Crawl-delay example for ClaudeBot; Google does not use the directive at all and crawl rate is handled through Search Console. If server load is a genuine problem, the answer is rate limiting on the server, not a line in robots.txt.

If I unblock a bot, how long until it comes back?

The decision is reversible, but not instantly. First the cache has to expire — generally 24 hours at Google, and OpenAI states roughly 24 hours. After that it depends on how often the crawler visits your site at all, which on a small, rarely updated site can take weeks. Blocking is fast; coming back is slow.

Is there a tool for testing robots.txt?

The robots.txt report in Google Search Console tells you how Google sees your file, but only for Google's crawlers. For the AI crawlers as a whole, Galata Media's free AI crawler access check interprets the file under the group-matching and longest-match rules and returns a verdict for sixteen bots. Used together, the two leave no blind spot.

Is llms.txt required?

No. llms.txt is a proposal, not a standard that search engines or model providers have jointly adopted, and its presence alone brings neither ranking nor citations. It is cheap enough to be worth publishing, but it sits behind crawler access and page structure in any order of priority.

Does llms.txt help SEO?

It has no direct effect on classic search ranking, and Google says so in its documentation: appearing in AI features requires no new machine-readable file. The indirect benefit is real though — writing the file forces a review of which pages actually explain anything, and that review usually ends in fixing the pages themselves.

How long should llms.txt be?

The specification sets no limit, but the file's purpose sets one for you: letting a model understand the site in a single read. On a corporate site that usually means 20–60 links. An llms.txt that runs to hundreds of lines has become a directory rather than a map, and reintroduces the problem the file was meant to solve.

I published llms.txt but still do not appear in ChatGPT — why?

Because the two are not connected. When ChatGPT answers a question it searches, and it searches with OAI-SearchBot; if that crawler is blocked in robots.txt or dropped by bot protection in front of the server, it never reads your site at all. llms.txt changes no link in that chain. Verify access first, then make sure the page carries a quotable answer passage.

How often should llms.txt be updated?

If the file is hand-written, it needs review after every new section or removed page — in practice, a monthly check. If it is generated from the content, the question answers itself, and that is the route to prefer. The cheapest control is a monthly script that requests every link in the file and reports the ones returning 404.

Does markup raise rankings directly?

No, it is not a direct ranking factor. What it does is remove ambiguity about what the page is about and make it eligible for rich result formats. On the generative side the effect is more concrete: markup turns the organisation from a string of text into a resolvable entity, and an engine does not recommend an entity it cannot resolve.

Is adding it with a plugin enough?

It is enough to start, but it has two limits. First, most plugins drop their own Organization block on every page and do not wire the nodes together with @id; the result is valid but disconnected. Second, if two plugins emit markup at once the page ends up with two contradictory records. Opening the page source after setup and counting the blocks is a five-minute check that pays for itself.

How should markup be written on a multilingual site?

Each language's page carries its own WebPage node with inLanguage set to that page's language, and its text fields — name, description — written in that language. The Organization node stays single: a company's name and address do not change by language. The link between language versions is made with hreflang, and there is no need to repeat it in the markup.

Which type should I start with?

Start with Organization and get it right: name, address, phone, logo, and a sameAs list of verifiable profiles. That single block is the node every page on the site will attach to; if it is wrong, everything built on it is wrong. Step two is a WebPage on every page, step three the type that fits — Service, Product or Article.

Can you submit a site to ChatGPT?

No — there is no submission form and no directory listing. Visibility depends entirely on crawler access and page structure. A service claiming to 'get you added to ChatGPT' should not be bought; the only sellable work is opening access and making the content quotable.

If I block GPTBot, do I also vanish from ChatGPT search?

No. GPTBot is for model training only. When ChatGPT searches in order to answer, it uses OAI-SearchBot, and when a user opens a link it uses ChatGPT-User. With those two open you can appear as a source in answers while GPTBot stays blocked.

Why does ChatGPT show my competitor and not me?

It is one of three reasons, checked in order. First access: the competitor's page can be crawled while yours is being dropped at the firewall. Second the passage: their page has a standalone paragraph answering the question directly, yours has marketing copy. Third the entity: the competitor is referenced more consistently in third-party sources.

Is ChatGPT sending traffic to my site, and how would I see it?

Yes — it arrives when a user clicks a link in the answer, and shows up in analytics as a referral from chatgpt.com. The volume is small next to classic search, but the intent is markedly more mature: the comparison already happened inside the model, and the visit lands at the decision stage. It is worth tracking as its own segment.

How long before results show?

Fixing access pays off fastest: once the cache expires the crawler returns, and that is measured in days. Passage and markup work reaching the answers depends on the page being recrawled and reprocessed; in practice the band runs four to twelve weeks. No supplier can honestly give a firm calendar for it.

Is there a way to opt out of AI Overviews?

There is no separate opt-out on the site side. Google states that AI is built into Search and that the control available to site owners is the robots.txt directives written for Googlebot — so leaving the summary entirely means leaving search. For limiting the text shown, the documented route is the nosnippet, max-snippet and data-nosnippet controls.

What is the difference between AI Overviews and AI Mode?

AI Overviews is the summary appearing above the classic results page; AI Mode is a separate experience where the whole search runs as a conversation. From a site owner's perspective the same rules govern both: both are fed by the search index, neither has additional requirements or special markup, and the data for both is folded into the same search type in Search Console.

My page is indexed — why does it never appear in the summary?

Being indexed is necessary but not sufficient. The summary is built from passages that answer the query directly; if your page discusses the topic without ever answering the question cleanly, there is no passage to take. The second common reason is the query being split into subtopics: being strong on the main topic does not guarantee being strong on the sub-questions fan-out produces.

Does being cited in the summary affect ranking?

No direct ranking reward is documented, and a claim like that would be unsourced. What is measurable is something else: a brand named inside the summary becomes visible at the very top of the results page, and that visibility works independently of its classic ranking position. The gain is not in a ranking score but in where attention lands.

What happens if I copy the manufacturer's description?

The page is not penalised, but it becomes indistinguishable. Among hundreds of retailers publishing the same text there is no reason for an engine to pick you; at that point the choice shifts to price, brand recognition and how complete the technical data is. The dimension, usage and compatibility detail you write yourself is the only thing separating your page from the rest.

Do I have to show the price in the markup?

If you want merchant listing experiences, yes: price and currency are required fields and the price must be greater than zero. In a model that does not publish prices — quote-based sales, dealer pricing — Product markup is still written, but merchant listings should not be expected. There, the product page's job is to be comparable on technical data instead of price.

Should filter pages be indexed?

Most should not. The product of colour, size and price-range filters generates thousands of addresses, nearly all of them marginal copies of one another. The rule: filter combinations with genuine search demand — "black office chair", say — get their own page and are indexed; the rest are tied back to the main category with a canonical.

Can we publish our own scoring as aggregateRating?

No. A rating in the markup must be visible on the page and must come from genuine customer reviews. A score a company awards its own product is an unverifiable claim; markup like that carries an enforcement risk and, on the generative side, drags down the credibility of the whole page.