Press review 22.11.2026: category description length, bookings through an AI assistant, an SEO plugin with a record number of flaws

Category pages with a description of 7,000-12,000 characters average 3 phrases in the top 3 and 340 monthly visits. Without a description: zero top-3 phrases and 11 visits.

Share
Press review 22.11.2026: category description length, bookings through an AI assistant, an SEO plugin with a record number of flaws

Topics that clearly more independent sources have covered in the last two weeks than before:

A short category description costs you traffic - here are the numbers from 204,000 pages

A category page with no description has no business in the top 3. In a study of 204,000 category pages such URLs averaged zero phrases in the top 3, one phrase in the top 10 and 11 visits a month.

Descriptions of 7,000 to 12,000 characters perform best. In that range a page picks up an average of 3 phrases in the top 3, 9-11 phrases in the top 10 and around 340 visits a month. Above 12,000-13,000 characters results start to fall, and at 15,000 the drop is clear.

The reason is simple: a very long text stops answering purchase intent and starts answering informational intent. The second one rarely wins queries about buying.

What to do: measure the length of category descriptions on the three pages you care about most and write the ones below 7,000 characters.

Sources:

Search Console API now separates branded queries from the rest - get the report automatically

Search Console added a brandedClassification dimension to its API. Reports can now split queries that contain the brand name from everything else without your own regular expressions.

There is one catch: data for this dimension only goes back to 21 February 2026. Older periods still need the old methods, and Google admits some queries may be classified incorrectly.

What to do: where the two methods overlap, compare the branded split from the API with your own filter before you hand the client a single number.

Sources:

Search Console will show Google Lens and Circle to Search results separately

A new search type called webMultimodal has appeared in the Search Console API. It returns results from Google Lens, from Circle to Search on Android and from image uploads.

Until now all of that traffic disappeared into regular web search. Now you can see how many people reach your site through a photo rather than a typed word.

What to do: if you sell products or visual services, add this type to your report and check whether the images on your site have any chance of being recognised.

Sources:

Rank Math has collected 16 vulnerabilities this year - check the version on every WordPress site

The Rank Math SEO plugin has collected 16 vulnerabilities this year. That is a record among SEO plugins: Yoast had 4 in the free version and 2 in the paid one over the same period.

Two of the flaws allowed remote code execution with Contributor or Author permissions. Five could be exploited without logging in, and the most serious scored 8.8 on the CVSS scale.

Affected versions are 1.0.279 and older. Eleven of the sixteen were discovered within seven weeks this autumn, so this is not a one-off mistake.

What to do: go through your clients' sites and check the Rank Math version, then limit Contributor and Author roles to the people who really need them.

Sources:

A new CDN with a cold cache and a bot block will cut your site out of the index

After switching to a new CDN the cache is empty. The first visit to every URL has to be served by your own server, and search engines visit far more URLs than people do.

Server load grows and the temptation appears to switch on bot protection. Googlebot cannot pass a CAPTCHA, so after a longer block pages start dropping out of the index.

Google recommends the opposite order: warm the cache first by fetching your most important URLs through the new CDN, then let the robots in. If the origin is overloaded, rate limiting for Googlebot beats a blanket block.

What to do: add a cache warm-up step and a check of verified bots to your CDN migration plan.

Sources:

Schema now has to explain your whole business to AI, not just a single page

Until now structured data had one job: give a page a chance at a rich result in Google. It now has a second role - explaining to models what the business is and how its parts connect.

An example from finance: an AI Overview marked a bank branch as permanently closed. Only strengthening the structured data helped correct the answer.

The authors point out that this is often not a hallucination but conflicting information the company itself published in different places.

What to do: take one service page and check whether the schema describes the same entity as the visible content - same address, name and specialisation.

Sources:

In AI Max you tell Microsoft which pages it may use - without blocking the rest

Signal: 2 independent sources this week, 3 in 30 days; tracked since 2026-07-21.

Microsoft Advertising added URL inclusions to AI Max. You point to addresses the system may use in three ways: as a traffic destination, as a source of ad copy and as a base for matching queries.

This is not the same as exclusions. Inclusions say what is welcome, not that traffic will go only there. If you want to cut sections out, you still need explicit exclusions.

The feature is aimed at large sites with many categories and changing promotions. Microsoft has published neither benchmarks nor a date for rolling it out to all accounts.

What to do: in a seasonal campaign point to your promotion pages and check in the reports whether paid traffic actually lands there.

Sources:

Local services ads are moving into Google Ads in waves - save your reports before they go

Google is not shutting down local services ads, it is moving them into Google Ads as a separate pay-per-lead campaign type. Most home service trades are already in the first wave.

The new campaigns still show in search and in Google Maps. That is the difference from a regular Performance Max campaign, which also runs outside those places.

The problem is in the data: report history from the old system does not move to the new interface.

What to do: before the migration, save the history and exports from the old panel, because after the switch there will be nothing to compare against.

Sources:

AI and tools

Google's assistant is starting to call businesses and book appointments for the customer

Google has released Gemini Tasks, an assistant that works like a personal agent. It sees your Gmail, calendar, Drive, Google Photos and Maps, and on that basis carries out tasks without asking about every step.

One feature concerns local businesses: the agent finds them near the user and books appointments. It can also pass on attachments from emails and tidy up the inbox. Agent shopping has a $1,000 cap and for now runs only in the US.

The feature is free and rolling out to more countries. For a service business it means the first contact will increasingly be an agent rather than a person.

What to do: check that the hours, phone number and form in your Google business profile are up to date - the agent will rely on exactly those.

Sources:

The same text written like an aircraft manual loses the waffle - one instruction does it

Aircraft maintenance manuals have one trait: nothing gets lost in them. Short sentences, one idea per sentence, no decoration. That rule is called ASD-STE100 and it has been circulating as a prompting trick for a while.

In practice the prompt tells the model to strip filler phrases and jargon and to break long sentences so each carries one thought. The meaning has to stay unchanged.

The result is often better than with "write it shorter", because the model gets a concrete pattern instead of a vague request for brevity.

What to do: run one product description through that prompt and compare it with the version on your site today.

Sources:

Claude Design shuts down on 14 December - move your projects yourself, some data will not follow

The Claude Design platform closes on 14 December. Its features - canvas layouts, interactive prototypes, one-page sites - move into the artifacts section of Claude itself as separate templates: design, slides and motion.

Projects and design systems can be moved manually through a banner in artifacts or from the project menu. Chats, comments, public links and email invitations will not move, so anything you want to keep has to be saved separately.

After 14 December the old platform becomes read-only. A finished artifact can be shown to anyone by link, without a Claude account.

What to do: if you keep client prototypes there, move them before 14 December and export the chat history, because afterwards it will only be a preview.

Sources:

Marketing, sales and business

An offer people feel stupid refusing has four variables - budget is not one of them

The value of an offer breaks down into four elements: the customer's dream outcome, their belief they can reach it, the waiting time, and the effort plus what they have to give up. Scarcity and urgency only amplify those four.

An example from the conversation: one company narrowed its service to the 15 percent of customers who bought the most expensive package and stayed the longest. After rebuilding the offer around that group it raised prices tenfold.

The speaker's conclusion: when customers are scarce, you usually need a better offer rather than a bigger marketing budget.

What to do: write down those four variables for your service and check where the customer waits longest and how much they have to do themselves.

Sources:

Add case studies to your bid and you can win without the lowest price

In a tender you do not have to be the cheapest. You can win on speed, on ease or on lower risk - and that is a choice worth making deliberately before you send the proposal.

The practical advice from the conversation: add one to three case studies at the end of your RFP response. They should show where the client was not the cheapest but saved the most. Then your own client does the selling, not your promises.

The example is a services firm with $18 million in revenue and retention above 95 percent that does almost no marketing and lives off tenders.

What to do: pick two completed projects with a measurable result and turn them into one-page case studies for your proposals.

Sources:

Annual billing instead of monthly lifts customer value fivefold

The difference between monthly and annual billing is not cosmetic. In data from 14,000 subscriptions the monthly plan lost an average of 10.7 percent of customers per month, quarterly 5 percent and annual 2 percent.

Going from 10 percent churn to 2 percent gives roughly five times the lifetime value of a customer. Even if only one in ten customers picks the annual plan, upfront cash grows noticeably.

The second piece of advice from the conversation: fix service delivery and retention first, and only then raise the price.

What to do: calculate churn in your billing model and check whether you offer an annual option with a concrete benefit.

Sources:

AI agrees with you about half again as often as a human - keep decisions away from it

An AI model agrees with the person talking to it around 49 percent more often than an ordinary human given the same data. That is reason enough not to use it for strategic decisions - you get confirmation of the direction you were already taking.

A second observation from the conversation: if profit does not grow after aggressive AI adoption, you are automating everything in turn rather than what actually blocks the business.

The author also argues that AI does not take risk, and pay follows risk. Someone has to keep responsibility for the outcome.

What to do: before your next agent deployment, name one constraint in the business it is supposed to move - and measure the effect after a month.

Sources: