How to audit your AI reputation.

Begin with one named entity and one decision context. A founder preparing for investor due diligence needs a different question set from a company trying to understand how buyers compare its services. Auditing both at once without separating the records makes errors harder to locate.

• For a leader: record the public name, spelling variants, current role, former roles, organizations, dates and areas of documented expertise.
• For an organization: record the brand and legal names, products or services, operating markets, official website and relevant leadership relationships.
• Choose the language, market and audience for this baseline. Test another language as a separate set rather than combining its results into one percentage.
• Build a reference sheet from current official records and independent sources. Mark unresolved claims as unverified; do not treat the entity’s own marketing as independent corroboration.

This reference sheet is the factual comparison point. The broader disciplines are explained in AI reputation and entity optimization.

Separate questions about the named entity from category questions that do not mention it. The first tests representation; the second tests whether the entity appears when a user is exploring options. Keep comparison questions distinct because they ask for a judgment, not simply a factual description.

One possible starting set is 30 prompts: 12 about the named entity, 12 about the category and six comparisons. This is a proposed working allocation, not a statistically validated sample size. Replace generic prompts with questions obtained from actual buyer conversations, search queries or stakeholder interviews.

Keep the exact wording and label every prompt with its audience and purpose. The prompt library provides templates; the research methodology explains the reporting standards.

Record the product, displayed model or mode when available, date and time, language, market assumptions, account context and whether search was enabled or actually used. An assistant answering from model knowledge and one retrieving current web pages are different test conditions.

• Use a fresh conversation for each independent test and keep the prompt wording unchanged across repeated runs.
• Save the complete answer, linked source URLs and a screenshot or export where available. A paraphrase alone is not an adequate audit trail.
• Repeat the prompt under the same documented conditions. Record variation rather than selecting the most favorable answer.
• Log refusals, failed runs, unavailable features and answers that do not address the question. State which observations are excluded from each calculation.
• Use public or approved information. Do not put confidential client records into consumer prompts merely to test whether an assistant repeats them.

For illustration, 30 prompts across three products, repeated three times, produces 270 observations before any additional languages or markets. Scope the workload before collection. This is an arithmetic planning example, not a measurement conducted for this article.

Download the blank audit log to record conditions, evidence and follow-up actions. It contains column headings only, with no client data or invented observations.

For every rate, publish both the count and the denominator. Mention rate can mean answers mentioning the entity divided by eligible answers in the defined set. Linked citation rate can mean answers linking an approved entity-owned domain divided by eligible answers. Specify the domain rule and avoid combining these with a competitive share-of-voice formula.

For accuracy, specify whether you checked claims or whole answers. State how many claims were checked and how many remained unresolved. The AI Share of Voice guide explains why relative share and presence rate answer different questions.

The following numbers are fictional and demonstrate reporting only. They are not client results, a provider benchmark or evidence that one engine performs better. Imagine five prompts repeated twice in each of two unnamed systems: ten eligible answers per system.

In this example, System A mentions the entity in seven answers, but only four answers link its owned domain. Those outcomes should not be reported as the same metric. The accuracy counts use a different denominator: the answers mentioning the entity. That definition must be visible beside the result.

The useful next question concerns the specific error and the source behind it. A higher mention rate does not compensate for a wrong current role, an incorrect service description or confusion with a different organization.

For each material discrepancy, record the claim, the evidence showing the problem, the linked sources if any, its likely business consequence and who can act. A cited page may explain an error, but a citation alone does not prove which passage caused the model’s wording. Keep that distinction in the audit.

• Address identity confusion and materially incorrect roles, services or dates first when they affect a decision the audience is making.
• Correct outdated or inconsistent information on the website and profiles you control. Preserve a clear record of the current fact and the date it changed.
• For independent sources, request a documented correction through the publisher’s normal process. Record the request and outcome; do not assume the correction will be accepted.
• Where a provider offers a relevant feedback or correction channel, retain the submission and its status. A report is not evidence that the answer has been changed.
• Give every action an owner, a source URL, a completion criterion and a retest condition. Distinguish a source correction from a subsequent change in an AI answer.

The diagnostic guides on inaccurate AI answers and entity confusion describe these situations in more detail.

Inspect the pages that should support accurate information: HTTP status, crawl access, indexing directives, declared canonical URL, internal links and visible text. Check that important facts appear in the page the visitor can read, rather than only in scripts or metadata.

Google states that its established SEO practices also apply to AI Overviews and AI Mode; it does not prescribe a special AI file for inclusion. Use Google’s AI features guidance to frame this part of the technical review. Crawl access is a prerequisite to investigate, not proof that a page will be selected.

OpenAI documents different purposes for OAI-SearchBot and GPTBot. Its crawler documentation should be checked when reviewing access: controls for search discovery and model training should not be treated as interchangeable.

Structured data should describe the visible article and its real author. A Markdown copy can offer another readable representation of the same content. Neither markup nor Markdown can instruct an independent AI system to adopt a claim or guarantee a citation.

• Entity reference sheet: verified facts, source links, dates and unresolved points.
• Prompt register and evidence log: exact questions, run conditions, answers and source URLs.
• Results by system, language and question family: counts, denominators and limitations.
• Source map: entity-owned pages, independent references, conflicting records and gaps in corroboration.
• Prioritized action list: issue, evidence, responsible owner, action and verification condition.
• Retest plan: stable prompts, comparable conditions and a dated change log.

Repeat the baseline after relevant source changes and indexing checks, using the same definitions. Document product or model changes as possible confounders. An observed improvement after an action does not, by itself, establish that the action caused it.

Keep the executive and company records connected but distinguish their results. The objective is a defensible view of representation and useful next actions. Any later claims about impact should be supported by the collected evidence and, where available, qualified enquiries and commercial outcomes.

This guide publishes a protocol and a blank audit log, not a completed study of Moroccan agencies or an agency ranking. No collected agency panel, saved test outputs or comparative findings are reported here. A study title is not evidence that a study has been conducted.

To investigate agencies in Morocco or another market, first declare the services, audience, cities or service areas, language, collection period and inclusion criteria. Keep AI consulting, SEO, web development, digital marketing and automation categories distinct when they answer different buyer needs. Verify the public identity and documented scope of each included organization.

Use the same reviewed question families for each eligible organization and record the exact wording, product, search context and repetitions. Category questions should not name the preferred agency; named-entity questions and comparisons belong in separate groups. Keep refusals, failed runs and unavailable features in the record with explicit exclusions.

Before publishing any findings, check each material claim against its linked source and report counts, denominators, uncertainty and collection conditions. Disclose selection rules, sponsorship and material relationships. The publisher offers related assessment services; this protocol is not an independent endorsement or ranking of providers.

Publish only conclusions supported by the actual records. A panel selected for a diagnostic is not automatically representative of every agency or buyer. Do not treat a missing mention as evidence of poor service, and do not turn citation counts into proof of commercial quality.

Use the existing blank audit log for authorized observations and the multilingual review guide when testing across languages or markets. Review the platform-versus-advisor comparison if tools or responsibilities need clarification. No fixed improvement timeline or guaranteed citation follows from this method.

What is an AI reputation audit?

An AI reputation audit is a documented review of how AI systems describe, cite and recommend a person or organization. It checks identity, factual claims, linked sources and competitive context under stated test conditions, then turns the findings into prioritized actions.

How is an AI reputation audit different from an SEO audit?

An SEO audit examines a website’s discoverability and search foundations. An AI reputation audit also checks the entity described in generated answers, including identity confusion, factual accuracy, independent sources and recommendations. The two audits share technical checks but have different scopes.

How many prompts are enough for an initial audit?

There is no universal number. Match the prompt set to the entity, audience, language, market and decisions being tested. A 30-prompt set can be a practical starting scope, but its size alone does not make the sample representative or validate an overall score.

Does publishing Markdown make AI cite my website?

No. A Markdown copy can make the same content available in another readable format, but publishing it does not establish that an AI system will retrieve, trust or cite the page. Accuracy, accessible pages and credible supporting sources still need to be assessed.

Can an audit guarantee that an AI answer will change?

No. An audit can identify observed errors, conflicting sources and actions within the entity’s control. It cannot guarantee a provider’s wording, source selection, retraining schedule or recommendation. Changes must be checked in subsequent recorded tests.

Does this guide publish a completed agency study or ranking?

No. It publishes a proposed protocol and a blank audit log, not collected findings or an agency ranking. A completed study would need declared inclusion rules, verified organizations, saved answers and sources, collection conditions, counts, denominators and limitations. No agency visibility result is reported here.

This article presents a proposed working method. Its numerical examples are fictional and do not report client results or a completed engine benchmark. Provider documentation was checked on 27 October 2026. Recheck it when planning an audit.

• Google Search Central — AI features and your website: Technical eligibility and the role of established SEO practices.
• OpenAI — Overview of OpenAI crawlers: Crawler purposes and access controls.
• Google Search Central — Article structured data: Markup that describes an article and its authorship.
• Mohammed Teto — Research methodology: The reporting framework used on this site; not evidence of a completed benchmark.

Related guidance: AI search intelligence, executive reputation and corporate AI reputation.