Search and AI briefing: ten checks for publishers

A revised August briefing on previews, citations, shopping data and publisher controls, with current documentation and clear verification steps.

Sonar untangles cards representing a control, an observation and an unanswered question.

This briefing reviews ten search and AI topics that affect publishers. It explains which settings you can change, which results you can check, and where you need more evidence. The aim is to help you choose a useful next step without assuming that a setting or tool can guarantee visibility.

First published on August 9, 2026, this briefing brings together reports reviewed during August 3–9. Some announcements were published earlier. The sources were checked again on November 5, and the guidance below reflects that review.

Brave’s benchmark: review the method before citing the result

Brave published its Search API announcement on February 12, 2026, and updated it on June 25. Its initial comparison used 1,500 queries, with answers judged pairwise by Claude Opus 4.5 and Claude Sonnet 4.5. The company evaluated each pair in both orders to reduce position bias. These are historical benchmark details, not an August launch.

Before citing a performance claim, check which test it describes. Record the queries, collection date, product versions, account conditions, and judging method. Brave collected its own answers directly and competitors’ answers through Bright Data. Consider whether those collection methods produced comparable results.

Brave reported these results; SearchEngineAnswer has not independently reproduced them. Reversing the answer order reduces one source of bias, but it does not show whether human reviewers would agree with the judges. Assess factual accuracy, supporting sources, and response speed separately before choosing a retrieval service.

DuckDuckGo previews: verify the current display

An earlier version of this briefing described expanded text summaries and playable video previews from DuckDuckGo’s quarterly updates. We could not confirm the archived Q2 details on the current updates page during this review. The timing remains unverified, and this briefing does not recommend special markup to obtain those previews.

To check your own pages, choose a small set of relevant queries. Record the country, device, date, displayed summary, and destination URL. If a video preview appears, check whether it accurately represents the video and whether the video plays on the destination page. Record missing previews as well. One search result cannot establish a general selection rule.

ChatGPT restaurant search: check account and provider eligibility

OpenAI’s current Search documentation lists reservation search on eligible consumer plans using GPT-5.6 Sol or GPT-5.6 Luna. The feature is not available in ChatGPT Work or on Business, Enterprise, or Edu plans. It lists OpenTable availability globally, Resy in the United States, and Yelp in the United States and Canada, subject to restaurant and location coverage.

If the Reserve option is missing, check whether the account is eligible and whether a supported provider lists the correct restaurant. Then check availability for the requested date, time, and party size. Depending on the provider, the booking may be completed in ChatGPT or on the provider’s website.

The reservation provider sends the confirmation and handles later changes or cancellations. ChatGPT does not save or manage existing bookings. Before treating a missing reservation as an SEO problem, check whether the issue concerns the restaurant listing, available tables, or the booking process.

ChatGPT shopping: distinguish product facts from generated labels

OpenAI’s shopping documentation says ChatGPT may simplify product titles and generate feature labels or review summaries from available information. A “Budget-friendly” label is not a verified award or proof of the lowest market price. OpenAI does not verify the displayed reviews and ratings.

If a result is wrong, compare the exact product variant, price, availability, and seller with the merchant’s current records. Check whether the problem comes from inaccurate source data, a delayed update, the wrong product variant, or a misleading generated summary. The right correction depends on the cause.

Shopify Catalog integration and direct product feeds are separate paths. Use the product-feed guide to review those paths without confusing organic product discovery, advertising, and checkout. Do not rewrite a factual product description to imitate a generated badge.

Google spam reports: submit evidence and protect private information

Google’s quality-report guidance says it may use spam reports to take manual action. It also says submission text must be sent to the affected site owner if an action is issued, and that reports containing personally identifying information may not be processed.

An effective report identifies observable behavior and the relevant policy. For example, record representative URLs and the redirect or doorway pattern a reviewer can reproduce. Remove personal details from descriptions and query strings. A competitor ranking above your site is not, by itself, evidence of a violation.

Google reviews the report and decides whether to act. Submitting it does not guarantee a response, a penalty, or a ranking change. A later change in search performance is not proof that the report caused it.

Google preview controls: choose the intended scope

Google’s robots meta documentation distinguishes three useful controls. nosnippet Suppresses text snippets and video previews and prevents page content from serving as direct input to AI Overviews and AI Mode. A static image may still appear. max-snippet Limits textual snippet length and the amount available as direct input, with documented exceptions for separately permitted uses. data-nosnippet Excludes selected portions from snippets.

Suppose an article contains an explanation followed by a promotional offer. Decide whether you want to restrict previews of the entire page or only the offer. A rule that applies to the whole page may limit more content than you intended. If you use a section-level attribute, check that it surrounds the intended passage in the rendered page.

Before changing a directive, check the page’s HTML, rendered content, and HTTP headers for conflicting instructions. These controls do not make a public page confidential. Training preferences and Google’s separate generative-AI inclusion setting require their own review; see the property-inheritance guide.

Perplexity citations: update how your application reads responses

Perplexity’s migration overview explains that Sonar accepts a messages array and returns choices. The Agent API accepts input and returns a typed output array that can contain search results and answer messages. Your application should identify the answer message instead of assuming that the first item contains it.

The migration procedure documents numbered citations for the fast effort level and source-typed citations for low, medium, and high. If your application recognizes only one citation format, test the other formats before migrating.

Prepare test cases for a claim with a supporting source, repeated citations, different source types, an empty result, and a failed retrieval. Keep each original citation marker, source identifier, URL, and associated claim. These are recommended tests; SearchEngineAnswer has not performed an API migration for this briefing.

Google images: use consistent metadata and check image crops

Google’s image guidance says selection is automated. Preferred-image metadata, including supported structured data and og:image, can influence the choice but does not force it. Make sure the metadata points to an image that represents the page and is accessible to crawlers.

For Discover, Google recommends relevant images at least 1,200 pixels wide. Enable large previews with max-image-preview:large or AMP. Meeting these requirements does not guarantee that the page will appear, as Google’s Discover guidance explains.

Check the original image, its responsive sources, and the square and landscape crops your publication serves. If essential context disappears when cropped, repair the composition rather than assume metadata will solve it. Keep alt text focused on what the image explains.

FAQ rich results: update your reporting and keep useful answers

Google’s documentation changelog records that FAQ rich results stopped appearing on May 7, 2026. This change predates the original August briefing. Don’t present the later removal of documentation as a second feature launch or a new loss of eligibility.

Review plugin settings, reports, and sales claims that still promise FAQ rich results in Google Search. Keep visible questions and answers when they help readers. The end of that search feature is not a reason to delete useful explanations or add repetitive content.

When reviewing performance around the change, mark the date in your reports. Check for changes in search demand, rankings, and page content as well as the loss of the FAQ display. Do not assume that every later change in clicks was caused by the markup.

SEO and GEO tool scores: identify what the number represents

Google’s third-party SEO guidance says external tools do not have access to its internal ranking data and cannot guarantee performance. A vendor score isn’t a Google measurement just because it uses a search-related label.

Ask what was collected, when it was collected, which pages or prompts were included, and how the inputs were weighted. Then identify the decision the score would change. A useful tool should show enough evidence for you to investigate its findings. An unexplained score threshold is not a reliable reason to change a page.

Compare a reported issue with a known page state before buying a remedy or rewriting content. Keep observed facts, estimates, and recommendations distinct. The measurement crosswalk helps prevent you from reporting a citation count, referral session, and ranking estimate as the same outcome.

Choose a check that informs a real decision

Start with a setting, page, or tool your publication already uses. Save its current state, define the expected result, and decide what evidence would justify a change. Before editing a preview directive, record which passage you want to restrict. Before migrating a citation parser, check that it passes the test cases.

This briefing is based on public documentation. It does not report an original benchmark or a live test of restaurant search, shopping results, or an API migration. The evidence-led publishing guide explains how to document your own test.

Revision note, November 5, 2026: Revised the wording, removed repeated introductions, and corrected the dates. Updated the restaurant eligibility and Perplexity guidance, and noted that the archived DuckDuckGo details could not be verified. The original publication date remains August 9, 2026.

Keep learning

Continue this topic

Community discussion

Discuss: Search and AI briefing: ten checks for publishers

Have a question, a useful example, or a different perspective? Join the discussion, share evidence, and help other readers reach a better answer.

0 replies Moderated
No replies yet.

Be the first to ask a focused question, share a practical example, or add useful evidence.

Ask a question or join the discussion

Share evidence, a useful example, or a clear question. Be specific, stay on topic, and challenge ideas without attacking people. First-time replies may be held for moderation.