ChatGPT SEO: publisher controls versus Google Search

Quick Answer: For ChatGPT SEO, check OAI-SearchBot permission in robots.txt; GPTBot controls training separately, and search-results adjustment can take approximately 24 hours.³ For Google's generative Search, check Search Console inclusion: worldwide from 31 August 2026, included by default, with parent-property inheritance where applicable.⁴ Neither control guarantees a citation.¹

Summary in a mind map

ChatGPT SEO: publisher controls versus Google Search
│
├─ Identify discovery
│   ├─ ChatGPT can rewrite questions for search partners
│   ├─ Google grounds answers in its Search index
│   └─ Test one customer question and intended URL
│
├─ Set OpenAI access
│   ├─ OAI-SearchBot governs search access
│   ├─ GPTBot concerns model training
│   └─ Search-results rules may take about 24 hours
│
├─ Check Google control
│   ├─ Search Console covers three AI feature groups
│   ├─ Worldwide since 31 August 2026
│   └─ Inclusion is default; parent choice can inherit
│
└─ Observe the result
    ├─ Record system, question, page and date
    ├─ Separate permission from citation
    └─ No publisher control guarantees placement

Start with the discovery path

A kitchen design showroom wants its guide to choosing worktop materials to appear when customers ask an AI assistant for options. The owner may call the work ChatGPT SEO. First, the team must identify the discovery system. Then it can edit the guide. OpenAI says ChatGPT can rewrite a question into targeted queries for search partners.¹ Google retrieves relevant pages from its Search index. Its generative features use core ranking systems.² The two explanations point to different paths that a publisher must assess separately.

This matters when a page is absent from an answer. The owner might assume the article needs a new AI format. But an absent citation can also arise because the page was unavailable to the relevant crawler, because the property was excluded from a feature, or because the answer system chose other sources. One prompt cannot reveal which of those occurred. The cited documentation offers no such simple inference. The task is to build a sequence of checks that isolates an access decision from a later source-selection decision.^1,2,3,4

The team begins with one customer question and one intended page. For the showroom, use a question such as which worktop material tolerates a busy family kitchen. Record the URL, the text that answers the question and the systems where the owner wants visibility. That small record prevents a later comparison from mixing different questions, pages and product features. It also makes the test repeatable. Someone else can run it after a settings change.

The table is a control map, not a claim that a permitted page will be selected. OpenAI explicitly declines to guarantee search placement.¹ Google's explanation of its grounding path does not turn every indexed page into a cited source.² The showroom should therefore track eligibility and observed citation as different fields.

Separate OpenAI's two bots

OpenAI's crawler overview distinguishes OAI-SearchBot from GPTBot. OAI-SearchBot is the permission relevant to search results; GPTBot concerns foundation-model training.³ The names can look like two labels for the same AI company, but they answer different publisher questions. A site owner should inspect the OAI-SearchBot rule for ChatGPT search. That rule addresses search access. A separate training preference belongs in the GPTBot row of the decision record.

The site team should read the actual robots.txt file served at the site's host and note the rule that applies to each crawler. It should not rely on a draft file in a repository or a remembered setting. Record when the file changed and the specific user agent affected. This is an operational check derived from OpenAI's separate crawler controls.³ It avoids the imprecise statement that a site is simply 'open to AI', which can conceal two distinct decisions.

For the showroom, a sensible worksheet has columns for host, file URL, OAI-SearchBot rule, GPTBot rule, time checked and person who approved the policy. The first two bot columns need independent entries. If the owner wants search access but a different training policy, the record should make that deliberate choice visible. The article does not prescribe a preference. It shows how to verify one using OpenAI's named controls.³

A robots.txt permission is still only one stage. It does not repair a missing page, an unclear answer or a poor match to a customer's question. OpenAI cites relevance and reliability factors. It does not guarantee placement in search results.¹ The team should therefore avoid interpreting a missing citation as proof that its crawler rule was ignored. It needs further evidence before changing the page or the permission.

Wait for crawler changes

Timing can distort a before-and-after test. For search results, OpenAI says adjustment to robots.txt changes can take approximately 24 hours, according to its crawler page as checked on 16 October 2026.³ That statement concerns search-results adjustment; it does not set a timetable for GPTBot training permission or promise a citation for the showroom's page.

Record the old and new rules. Note when the revised file was served. A test conducted immediately after the edit should be labelled provisional. Compare a later test with it. A missing citation after 24 hours does not, by itself, establish a crawler fault. OpenAI's placement caveat still applies.¹ That distinction prevents a permission change from becoming an unsupported promise to the marketing team.

Several same-day changes make the evidence harder to read. The showroom might alter robots.txt, rewrite its guide and change navigation. All three could occur at once. If the page appears in an answer later, the team cannot identify which change mattered. A narrow test has one permission decision and one dated content version. The method does not promise causality, but it keeps the record coherent enough to diagnose obvious access mistakes.

The resulting note should say: 'OAI-SearchBot permission changed at this time; OpenAI says adjustment can take about 24 hours; citation checks remain observations, not guarantees.' That is a more accurate statement than 'ChatGPT indexing completes tomorrow'. The cited source speaks to systems adjusting to robots.txt, not to a universal schedule for every URL.³

Read Google's site setting

Google's Search generative AI control sits in Search Console settings. Its help documentation says the choice covers AI Overviews, AI Mode and generative features in Discover.⁴ The site owner should identify the correct property before reading the choice. If the showroom uses a parent property and a more specific child property, the child may inherit its parent's setting by default. That value may reflect inheritance. It need not be a fresh local decision.⁴

The control became available worldwide on 31 August 2026, according to the cited Google help page as checked on 16 October 2026. Inclusion is the default; a property with a parent defaults to inheriting the parent's control.⁴ These points answer three questions. Is the feature available? What happens without a choice? How does a child property obtain its value? The person checking the page should note each separately. 'We did not change the setting' is not the same as 'the property is excluded'.

For the showroom, write down the Search Console property, any parent property, the displayed inclusion state and the time it was checked. Then identify the guide's URL that the team wants to assess. The setting concerns a property and a group of Google's generative features. It is not an OpenAI robots.txt rule and should not be copied into the ChatGPT column of the worksheet. Keeping the two systems in separate rows avoids a false single 'AI opt-in' status.^3,4

Google's guide also says its generative Search experience is grounded through pages from the Search index.² The inclusion control and ordinary search eligibility therefore answer different questions. A property may allow the feature, while a particular page still needs to be relevant and retrievable through Search. Record the property setting first. Then check the page itself. It should not report the property choice as a URL-level guarantee.

Compare defaults and scope

The two publishers' controls differ in where they live and what they cover. OpenAI describes crawler-specific permissions in robots.txt, with OAI-SearchBot and GPTBot serving different purposes.³ Google describes a Search Console inclusion choice for AI Overviews, AI Mode and generative features in Discover.⁴ These settings are not interchangeable. Changing the wrong one will miss the intended feature.

The OpenAI row intentionally avoids inventing a universal default robots.txt policy for a particular site. The actual file and applicable rule need inspection. The Google row reflects Google's published default and inheritance behaviour.⁴ The team should not flatten these two sorts of defaults into a single yes-or-no label. One is an access instruction found in a served file; the other is a product setting attached to a verified property.

Test one page and question

Choose the worktop guide as the first test case. Record the customer's question in its natural wording, the URL intended to answer it and the visible paragraph that contains the useful answer. A page about material choices should make its criteria explicit rather than relying on a gallery image. This is editorial advice for the showroom, not a claim that either company rewards a particular paragraph layout. Google's account of index-based grounding and OpenAI's relevance caveat make clarity worth checking without promising a ranking result.^1,2

Next, verify the controls in the order that can be observed. The team reads the served robots.txt file for OAI-SearchBot and GPTBot. It notes the change time if either rule was edited. It checks the correct Search Console property and whether the Google control is included, excluded or inherited. Finally, it records the page's ordinary search state and runs the chosen question in the experiences it cares about. The record keeps a control failure from being mistaken for a content judgement.^2,3,4

Repeat with the same question after a meaningful change. If the question changes, label it as a new test instead of blending it into a trend. Different runs may show different sources. That variation does not make the control record useless; it means the outcome must be stated as an observation. OpenAI's documentation explicitly prevents a guaranteed-placement conclusion.¹

Limits of a publisher choice

Neither published control can decide whether a business deserves a citation for every query. OpenAI says ChatGPT uses multiple relevance and reliability factors and does not guarantee placement.¹ Google explains a search-index grounding path but does not say every eligible page will support an AI response.² A control can make a page available. The response still depends on the question and source selection.

The cited pages also do not establish a rollout date for an unrelated change to Googlebot or AMP hosting. A search phrase may combine those ideas with the new Search Console setting, but the source for this article only supports the 31 August 2026 worldwide availability of the generative AI control.⁴ The showroom should avoid merging separate product announcements into one timeline. If it needs an AMP or crawler-specific operational decision, it needs documentation for that separate change before making a claim.

What to do next

Pick one customer question and its intended URL. Record OAI-SearchBot and GPTBot permissions, then check the relevant Google Search Console property's generative AI control and any inherited value. Note any OpenAI robots.txt change time. Test the same question after the approximate adjustment period, recording what appears without promising a citation.^1,3,4

The next reading depends on the diagnosis. Check whether ChatGPT recommends the business, how to observe brand mentions, what generative engine optimisation means and why a competitor may appear instead. These are related questions, not substitutes for the control check above.

Use the AI Answer Audit for the next review of how your brand appears in AI answers.

In words, the tree first asks whether the owner is checking ChatGPT or Google Search. The ChatGPT branch records separate OAI-SearchBot and GPTBot permissions and allows time for OpenAI's change to settle. The Google branch checks the Search Console property and its parent choice. Both branches end with a dated question-and-URL observation, not a guaranteed citation.